Piezoelectric Sensors for Chewing and Swallowing Detection: A Research and Development Guide

Abstract

This article provides a comprehensive overview of the use of piezoelectric sensor technology for the objective detection and analysis of chewing and swallowing. Aimed at researchers, scientists, and drug development professionals, it covers the foundational principles of piezoelectricity and swallowing physiology, explores current sensor designs and data processing methodologies, addresses key technical and optimization challenges, and reviews validation protocols and performance metrics. By synthesizing the latest research, this guide serves as a critical resource for advancing the development of non-invasive, reliable monitoring tools for dysphagia screening and pharmaceutical applications.

The Science of Swallowing and Piezoelectric Sensing

The process of swallowing, or deglutition, is a complex neuromuscular sequence that transports a bolus from the oral cavity to the stomach while protecting the airway. This sophisticated mechanism involves precisely coordinated interactions between sensory inputs and motor outputs across multiple anatomical structures. Traditional swallowing assessment methods often rely on subjective clinical observations or invasive instrumental procedures like videofluoroscopy, which pose limitations for continuous monitoring and early screening [1]. Recent technological advances have opened new avenues for objective, non-invasive measurement of swallowing function. Within this context, piezoelectric sensor technology has emerged as a promising tool for detecting and analyzing the physiological events of chewing and swallowing, offering the potential for automated, efficient, and highly sensitive assessment across clinical and research settings [2] [3]. These application notes detail the integration of this sensor technology with established swallowing physiology paradigms.

Physiological Stages of Swallowing

Swallowing is traditionally divided into four distinct yet continuous stages: the oral preparatory, oral, pharyngeal, and esophageal stages. Each stage involves specific physiological actions that can be detected and quantified using sensor-based technologies.

• Oral Preparatory Stage: This voluntary stage involves food intake, bolus formation, and lubrication. Mastication (chewing) reduces the food particle size and mixes it with saliva. The key detectable events are jaw movements during chewing and the compression of the bolus against the palate. Piezoelectric strain sensors placed below the ear can effectively capture these mandibular movements [2].
• Oral Stage: This is the voluntary transit of the bolus from the front to the back of the oral cavity. The tongue propels the bolus posteriorly towards the oropharynx. The end of this stage is marked by the triggering of the pharyngeal swallow, a critical event that can be identified by a distinct laryngeal elevation.
• Pharyngeal Stage: This involuntary stage begins with the triggering of the swallow reflex, which seals the airway and propels the bolus through the pharynx. Key events include:
  • Soft palate elevation to close the nasopharynx.
  • Laryngeal elevation and anterior movement.
  • Vocal fold adduction to protect the trachea.
  • Epiglottic inversion to cover the laryngeal vestibule.
  • Sequential pharyngeal constrictor muscle contraction.
  • Upper esophageal sphincter (UES) relaxation and opening. Acoustic sensors, such as a neck-worn electronic stethoscope (NWES) positioned over the laryngopharynx, are highly effective at detecting the characteristic sound of the bolus passing during this phase, as well as the associated laryngeal movement [4] [2].
• Esophageal Stage: This involuntary stage involves the transport of the bolus through the esophagus via peristaltic waves towards the stomach. The UES closes after bolus passage, and the lower esophageal sphincter relaxes to allow entry into the stomach. While this stage is more challenging to monitor with external sensors, vibratory signals may still be captured.

Sensor Technologies for Detection

The following table summarizes the key sensor technologies used for detecting events across the different stages of swallowing.

Table 1: Sensor Technologies for Swallowing Phase Detection

Swallowing Stage	Detectable Physiological Event	Primary Sensor Technology	Sensor Placement
Oral Preparatory	Jaw movement (Mastication)	Piezoelectric strain sensor [2]	Below the ear, on the cheek
Oral Preparatory	Bolus compression, tongue movement	Piezoelectric sensor in oral device [5]	Integrated into smart glasses (temple tip)
Pharyngeal	Laryngeal movement, Bolus passage sound	Neck-Worn Electronic Stethoscope (NWES) / Acoustic Sensor [4] [2]	Anterior neck, between C2-C5 vertebrae
Pharyngeal	Laryngeal excursion	Piezoelectric strain sensor [2]	Over the laryngopharynx (coniotomy region)

Application Notes: The Test of Masticating and Swallowing Solids (TOMASS) Protocol

The Test of Masticating and Swallowing Solids (TOMASS) is a validated screening tool that assesses the integrated process of mastication and swallowing of solid foods [4]. The conventional method relies on visual observation and manual timing, which introduces operator-dependent variability. The integration of a Neck-Worn Electronic Stethoscope (NWES) enables a semi-automated, objective assessment.

Key Quantitative Outcomes from NWES-TOMASS Validation

A recent pilot study with 123 healthy adults demonstrated the feasibility of using a NWES for automated TOMASS evaluation. The study reported the following key parameters, highlighting age and gender-related differences [4].

Table 2: Quantitative Parameters from a NWES-based TOMASS Study (n=123) [4]

Parameter	Description	Representative Median Values (IQR)	Statistical Findings
Discrete Bite Count	Number of bites to consume one cracker	Younger Women: 2.3 [1.0-3.0] vs. Younger Men: 1.0 [1.0-2.0]	Significant gender difference (p=0.042)
Swallow Count	Number of swallows to consume one cracker	Younger Women: 2.5 [2.0-2.5] vs. Younger Men: 2.0 [1.0-2.0]	Significant gender difference (p=0.026)
Oral Processing and Swallowing Time (OPST)	Time from first bite to completion	Prolonged with age, particularly in men (p<0.001)	Significant age-related change
First OPST (1st-OPST)	Time from first bite to first swallow	Prolonged with age, particularly in men (p<0.001)	Significant age-related change

Detailed Experimental Protocol: NWES-TOMASS

Objective: To objectively measure masticatory and swallowing performance during solid food consumption using a Neck-Worn Electronic Stethoscope (NWES).

Materials and Reagents: Table 3: Research Reagent Solutions and Essential Materials

Item	Function/Description	Specification / Example
Neck-Worn Electronic Stethoscope (NWES)	Primary sensor for automated detection of swallowing sounds via deep learning-based analysis of cervical auscultation signals [4].	Piezoelectric vibration sensor; positioned between C2-C5.
Smartphone with Data Acquisition App	Device for recording audio data from the NWES and capturing synchronized video footage [4].	e.g., Nexus 5X (Android); acts as data logger and sync hub.
Test Food	Standardized solid bolus for provoking a measurable masticatory and swallowing sequence [4].	Two Nabisco Premium Crackers (3g each, 47x47x3mm).
Data Annotation & Analysis Software	Software platform for manual integration, synchronization, and annotation of audio-video data to create a gold standard for algorithm validation [4] [5].	ELAN (Version 6.6 or later, Max Planck Institute for Psycholinguistics).

Procedure:

• Sensor Setup: The participant dons the NWES around the anterior neck, ensuring the sensor is positioned between the C2 and C5 vertebrae. The device is connected via a wired connection to a smartphone for data recording [4].
• Data Recording: The smartphone's video camera is activated to capture a close-up view of the participant's mouth and hand-to-mouth actions. The NWES is activated to record swallowing sound signals.
• Test Administration: The participant is provided with two separate crackers. They are instructed to eat one cracker at a time at their normal pace and to verbally indicate "Finished" upon complete swallowing of each cracker [4].
• Data Synchronization: The recorded audio data from the NWES and video data from the camera are manually integrated and synchronized using annotation software (e.g., ELAN). The audio waveform corresponding to the "Finished" utterance serves as the alignment point [4].
• Data Reduction and Analysis:
  • Discrete Bite Count: Manually counted by a trained human coder from the video recording.
  • Swallow Count: Counted using a combination of the synchronized video and the audio waveform from the NWES.
  • Oral Processing and Swallowing Time (OPST): The duration from the initial sound of biting the cracker (visible on the audio waveform) to the onset of the "Finished" utterance is measured.
  • First OPST (1st-OPST): The duration from the initial bite sound to the onset of the first swallowing sound is determined [4].

Experimental Workflow and Data Analysis

The following diagram illustrates the end-to-end workflow for a sensor-based swallowing assessment study, from participant preparation to data analysis and model validation.

The integration of piezoelectric and acoustic sensors into swallowing physiology research provides a robust, objective methodology for quantifying deglutition. The NWES-based TOMASS protocol exemplifies how operator-dependent limitations of conventional clinical screenings can be overcome, enabling the detection of subtle, clinically significant variations related to age and gender [4]. These technological advancements, particularly when employing multimodal sensing approaches, show significant promise for enhancing early dysphagia detection and long-term monitoring of at-risk populations [1]. Future research should focus on the external validation of these automated systems in diverse clinical populations and the expansion of these principles to monitor a wider range of ingestive behaviors, solidifying the role of sensor technology in the future of deglutition science.

Fundamental Principles of the Piezoelectric Effect

The piezoelectric effect is a fundamental electromechanical interaction describing the ability of certain materials to generate an electric charge in response to applied mechanical stress. This direct piezoelectric effect, discovered by Jacques and Pierre Curie in 1880, forms the basis for a wide range of sensing applications. The converse piezoelectric effect, wherein materials undergo mechanical deformation in response to an applied electric field, enables actuator functionality.

The underlying mechanism involves the displacement of ions within crystal structures that lack a center of symmetry. When mechanical stress distorts the crystal lattice, the resulting displacement of positive and negative charge centers creates a net electric dipole moment, manifesting as measurable electrical potential across the material. This linear electromechanical coupling makes piezoelectric materials exceptionally valuable for transducing mechanical phenomena—such as pressure, force, and acceleration—into quantifiable electrical signals.

Key Piezoelectric Materials and Properties

Piezoelectric materials are categorized by their composition and structure, each offering distinct advantages for specific applications. The performance of these materials is characterized by several key parameters, including piezoelectric charge constant (dij), piezoelectric voltage constant (gij), electromechanical coupling factor (k), and acoustic impedance (Z).

Table 1: Comparison of Key Piezoelectric Material Properties

Material Type	Example Materials	Piezoelectric Charge Constant (d33, pC/N)	Relative Flexibility	Biocompatibility	Primary Applications
Ceramics	Lead Zirconate Titanate (PZT)	300-650	Low (Brittle)	Poor (Contains Lead)	Ultrasound transducers, sensors, actuators
Polymers	Polyvinylidene Fluoride (PVDF)	20-30	High	Good	Wearable sensors, acoustic transducers
Polymer Nanocomposites	PVDF with piezoceramic particles	50-100	Moderate	Good	Structural health monitoring, flexible devices
Lead-Free Hybrids	Bismuth Iodide-based hybrids	Comparable to PZT	High	Excellent	Wearable technology, medical implants

Recent material advances have focused on addressing limitations of traditional piezoelectrics. Lead-free piezoelectric materials, such as those based on bismuth iodide, have emerged with efficiency comparable to commercial lead-based ceramics while offering reduced toxicity and lower processing temperatures [6]. Polymer-based piezoelectric nanocomposites provide enhanced flexibility, lightweight characteristics, and integration advantages compared to non-polymeric counterparts, making them particularly suitable for wearable technology and biomedical applications [7].

Piezoelectric composites (0-3 composites), typically consisting of a polymer matrix containing piezoceramic particles, offer unique benefits including lower acoustic impedance and the ability to detect high-frequency waves, making them valuable for structural health monitoring applications [8]. Their acoustic impedance is closer to biological tissues and carbon fiber-reinforced polymers than traditional piezoceramics, resulting in improved signal transmission and reduced signal reflection at material interfaces.

Piezoelectric Sensors for Chewing and Swallowing Detection: Application Protocols

The application of piezoelectric sensing to swallowing and mastication assessment represents a significant advancement in objective dysphagia screening. The following protocols detail methodology for utilizing piezoelectric sensors in this domain.

Neck-Worn Electronic Stethoscope for Swallowing Assessment

This protocol outlines the procedure for using a neck-worn electronic stethoscope (NWES) incorporating piezoelectric sensors to automatically detect and monitor swallowing actions through deep learning-based analysis of collected sound data [4].

Table 2: Research Reagent Solutions for Swallowing Assessment

Item	Specification	Function/Purpose
Neck-Worn Electronic Stethoscope (NWES)	Piezoelectric vibration sensor	Detection of swallowing sounds through cervical auscultation
Test Food	Two Nabisco Premium Crackers (3g each, 47×47×3mm)	Standardized solid bolus for TOMASS assessment
Data Acquisition Smartphone	Nexus 5X (Android 8.1.0) or equivalent	Records audio data via wired connection to NWES
Video Recording Device	Smartphone camera	Visual documentation of cracker consumption
Analysis Software	ELAN (Version 6.6, Max Planck Institute)	Synchronization and annotation of audio-video data

Experimental Procedure:

• Sensor Placement: Position the NWES around the anterior neck between the C2 and C5 vertebrae, ensuring proper skin contact for optimal signal acquisition.

• Equipment Setup: Connect the NWES to the smartphone via wired connection. Initiate both audio recording through the dedicated application and video recording using the smartphone camera to capture close-up footage of cracker consumption.

• Test Administration: Provide participants with two individual crackers sequentially. Instruct participants to eat one cracker at a time at their normal pace and to verbally indicate "Finished" upon completion of each cracker.

• Data Synchronization: Manually integrate recorded audio and video data by aligning the audio-video correspondence of the "Finished" utterance using analysis software (ELAN).

• Parameter Extraction:

  • Discrete Bite Count: Count the total number of bites required to consume a single cracker from video recordings.
  • Swallow Count: Determine the number of swallows required to finish a single cracker using combined video recordings and audio waveforms.
  • Oral Processing and Swallowing Time (OPST): Measure the duration from the initial sound of biting the cracker to the onset of verbal completion indication.
  • First Oral Processing and Swallowing Time (1st-OPST): Calculate the duration from the initial biting sound to the onset of the first swallow sound.

• Data Analysis: Employ statistical analysis (e.g., Kruskal-Wallis test, ANOVA) to assess age- and gender-related differences in TOMASS parameters.

OCOsense Glasses for Chewing Behavior Monitoring

This protocol describes the validation methodology for using piezoelectric sensor-equipped glasses to detect chewing behavior through facial muscle movements [5].

Experimental Procedure:

• Equipment Setup: Participants wear OCOsense glasses equipped with piezoelectric sensors to detect facial muscle movements during eating.

• Test Session: Conduct a 60-minute lab-based breakfast session while simultaneously recording behavior with video cameras for manual annotation.

• Data Collection: Record chewing data from specific foods (e.g., bagel and apple) simultaneously via both the OCOsense glasses' algorithm and manual behavioral coding.

• Validation Analysis:

  • Compare the number of chews recorded within each eating segment between manual coding and algorithm output.
  • Calculate correlation coefficients between the two measurement methods.
  • Assess mean chewing rates for significant differences between methods.
  • Evaluate participant self-assessment of eating rate against recorded chewing behavior.

• Performance Metrics: Determine the percentage of correctly detected eating and non-eating behavior episodes.

Experimental Workflow and Signal Pathways

The following diagrams illustrate the fundamental operational principles and experimental workflows for piezoelectric sensing in swallowing and chewing assessment.

Piezoelectric Sensing Pathway

Swallowing Assessment Protocol

Performance and Validation Metrics

Recent studies have demonstrated the effectiveness of piezoelectric sensing for chewing and swallowing assessment. The NWES approach has enabled objective TOMASS measurements, revealing age-related prolongation of OPST and 1st-OPST, particularly in men (p < 0.001), and gender-based differences in bite and swallow counts among younger adults [4]. OCOsense glasses validation showed strong agreement with manual video annotations, with no significant difference in chew count between methods and a strong correspondence (r(550) = 0.955) between manual coding and algorithm output [5].

In the broader context of dysphagia screening technologies, a systematic review of AI and sensor-based approaches reported performance varying widely with accuracy from 71.2% to 99%, area under the receiver operating characteristic curve ranging from 0.77 to 0.977, and sensitivity ranging from 63.6% to 100% [1]. Multimodal systems generally outperformed unimodal systems, highlighting the potential of piezoelectric sensors as components of integrated assessment approaches.

Implementation Considerations

Successful implementation of piezoelectric sensing for chewing and swallowing detection requires attention to several critical factors. Sensor encapsulation is essential for electrical insulation and protection from environmental factors, though it may slightly reduce sensitivity at lower frequencies [8]. Acoustic impedance matching between piezoelectric materials and biological tissues improves signal transmission efficiency, with piezoelectric composite sensors (PCS) offering advantages over traditional piezoceramics for biological applications [8].

Signal processing approaches increasingly incorporate artificial intelligence, with support vector machines being the most common model (62% of studies) and deep learning approaches emerging in recent years [1]. Future research directions include the development of lead-free piezoelectric materials with improved biocompatibility [6], integration with multimodal sensing systems [1] [9], and validation in diverse clinical populations including those with dysphagia and masticatory dysfunction [4].

Dysphagia, or swallowing difficulty, is a prevalent and serious medical condition that poses a significant threat to patient health and quality of life. Current epidemiological data reveals that dysphagia affects 8–22% of populations over 50 years old, with this number being generally higher for nursing home residents [10]. The condition is particularly widespread among individuals with neurological conditions, affecting 58% of older adults with dementia [1]. The clinical consequences of undiagnosed or poorly managed dysphagia are severe, including life-threatening complications such as aspiration pneumonia, malnutrition, and dehydration [11] [10]. Most alarming is the mortality association—patients with dysphagia experience a 13-fold higher risk of mortality compared to those without swallowing difficulties [1].

The current gold-standard diagnostic tools—videofluoroscopic swallowing study (VFSS) and fiberoptic endoscopic evaluation of swallowing (FEES)—though clinically valuable, present significant limitations that restrict accessibility [11] [10] [1]. These methods require specialized clinical settings, expensive equipment, and trained experts for operation and interpretation. Furthermore, they carry inherent risks including radiation exposure in the case of VFSS and the invasive nature of transnasal insertion for FEES [10] [12]. These limitations create substantial barriers to routine screening, early detection, and long-term monitoring of swallowing function, particularly in resource-limited settings or for home-based care.

Emerging Technological Solutions: Piezoelectric Sensors

Sensor Technologies for Dysphagia Assessment

Recent advancements in wearable sensor technology have opened new possibilities for non-invasive, accessible dysphagia screening. Among these, piezoelectric sensors have emerged as a particularly promising modality due to their ability to detect laryngeal movements through skin deformation during swallowing [11] [10]. The table below compares different sensing approaches mentioned in the literature:

Table 1: Sensing Technologies for Swallowing Assessment

Technology	Detection Principle	Advantages	Limitations
Piezoelectric Sensors [11] [10] [12]	Mechanical deformation of piezoelectric materials generates electrical charge	High sensitivity to laryngeal movement; flexible form factors; non-invasive	Signal interference from head movements; positioning critical
Accelerometers [11] [10]	Measurement of laryngeal acceleration during swallowing	Simple signal processing; well-established technology	Less specific to swallowing events; sensitive to motion artifacts
Acoustic Sensors [2] [1] [13]	Detection of swallowing sounds via microphone	High temporal resolution; non-contact options available	Background noise interference; privacy concerns
Surface EMG [11] [10]	Measurement of muscle electrical activity	Direct measurement of muscle activation patterns	Sensitive to skin impedance; requires conductive gel

Advanced Piezoelectric Materials

The development of novel piezoelectric materials has significantly enhanced the potential for wearable dysphagia monitoring. Recent research has focused on two primary material systems:

Aluminum Nitride (AlN) Thin Films: These sensors consist of a thin-film heterostructure with AIN (1 μm) as the piezoelectric layer, deposited on a flexible Kapton substrate (25 μm thick) [11]. This configuration creates an ultrathin, compliant patch (total thickness 26 μm, weight <2 g) that conforms to neck skin contours with minimal anatomical obstruction [11]. The biocompatibility of AIN and its compatibility with standard microfabrication techniques make it particularly suitable for biomedical applications [11].

Lead-Free Hybrid Materials: A groundbreaking development is the creation of a bismuth iodide-based organic-inorganic hybrid material that demonstrates piezoelectric performance comparable to traditional lead-based ceramics but without toxicity concerns [14] [6]. This material can be processed at room temperature (unlike PZT which requires ~1000°C) and offers a favorable combination of mechanical flexibility and strong piezoelectric response [14] [6]. The material's design leverages halogen bonding between organic and inorganic components to create structural instability that enhances piezoelectric performance [6].

Experimental Protocols for Dysphagia Sensor Evaluation

Sensor Fabrication and Characterization

AIN Piezoelectric Sensor Fabrication: The fabrication process employs standard microfabrication techniques including photolithography and sputtering deposition [11]. The layer structure consists of: (1) aluminum nitride interlayer (120 nm), (2) molybdenum bottom electrode (200 nm), (3) piezoelectric aluminum nitride layer (1 μm), and (4) molybdenum top electrode (200 nm) [11]. Kapton foil (25 μm thick) serves as the flexible substrate. An innovative 3D-printing system (DragonFly LDM, Nano Dimension) implements metal contacts on electric pads within the sealed package without affecting device performance [11].

Piezoelectric Sensor Array Construction: An alternative approach involves creating small piezo pressure sensors (1.5 mm length, 7.0 mm width) using polyvinylidene fluoride (PVDF) sheets (40 μm thickness) integrated into stainless steel cases [12]. Five sensors are aligned with 3.0-mm intervals and embedded in a urethane resin sheet (80 mm × 100 mm × 8 mm) with silicone gel between sensors to suppress interference [12]. The sensor surface is positioned slightly above the urethane sheet surface to ensure contact with skin during laryngeal movement [12].

Performance Evaluation Methodologies

Laryngeal Motion Simulator (LMS) Testing: A custom electromechanical setup mimics the diagonally upward and downward motions of the laryngeal prominence during swallowing [11]. The system uses a 3D-printed structure with a stepper motor (Mercury Motor, SM-42BYG011-25) controlled by an Arduino microcontroller unit to replicate laryngeal kinematics. An Ecoflex (Ecoflex 00-50, Smooth-on) membrane simulates the mechanical properties of neck skin [11]. The protruding part of the LMS ensures a reliable height (about 12 mm) of the simulated laryngeal prominence, consistent with literature values for normal swallowing [11].

Human Subject Validation: For clinical validation, subjects are seated upright with the sensor array lightly attached to the ventral surface of the neck near the laryngeal prominence [12]. The lowest-positioned sensor is placed 0.5-1.0 cm higher than the laryngeal prominence at rest [12]. Subjects are instructed to hold 3 ml of water in their mouth until instructed to swallow, with data collected over 10-20 swallowing trials per subject [12]. Simultaneous video recording of the neck (320 × 240 pixels at 30 frames/s) provides reference data [12].

The diagram below illustrates the experimental workflow for sensor validation:

Signal Processing and Data Analysis

Quantitative Swallowing Parameters: The following parameters can be extracted from piezoelectric sensor signals to characterize swallowing function [11] [12]:

• Swallowing Latency: Time from swallow initiation to completion (approximately 0.49-0.53 s in healthy subjects) [12]
• Maximum Rising Velocity: Speed of laryngeal elevation during swallowing (0.08-0.11 m/s in healthy subjects) [12]
• Maximum Lowering Velocity: Speed of laryngeal descent post-swallow (0.09-0.11 m/s in healthy subjects) [12]
• Swallowing Duration: Total time of the swallowing sequence [11]

Signal Processing Workflow: The diagram below illustrates the signal processing pathway from raw sensor data to clinical insights:

Research Reagent Solutions and Materials

Table 2: Essential Research Materials for Piezoelectric Dysphagia Sensor Development

Material/Component	Specifications	Function/Application
Piezoelectric Materials
Aluminum Nitride (AlN) [11]	1 μm thickness, sputter-deposited	Piezoelectric sensing layer for flexible sensors
Polyvinylidene Fluoride (PVDF) [12]	40 μm thickness, d₃₃: 35 pC/N	Flexible piezoelectric polymer for pressure sensing
Bismuth Iodide Hybrid [14] [6]	Organic-inorganic halobismuthate	Lead-free piezoelectric material for environmentally friendly sensors
Substrate Materials
Kapton Foil [11]	25 μm thickness	Flexible substrate for thin-film piezoelectric sensors
Urethane Resin Sheet [12]	80×100×8 mm, Asker C hardness: 5	Support matrix for sensor array integration
Silicone Gel [12]	θ-7, Asker C hardness: 0, 1.5 mm thickness	Sensor isolation and interference suppression
Electronic Components
Impedance Conversion Circuit [12]	Gain: 0.56, Time constant: 3.0 s	Signal conditioning for piezoelectric sensors
Microcontroller Unit [11]	Arduino Nano	Motion control for laryngeal movement simulator
Stepper Motor [11]	Mercury Motor, SM-42BYG011-25	Precision movement generation for LMS testing
Characterization Equipment
Single-Crystal X-ray Diffraction [14] [6]	N/A	Material structure analysis for novel piezoelectrics
Solid-State NMR [14] [6]	N/A	Local chemical environment characterization
Analog-Digital Converter [12]	Power1401-3, CED	High-frequency signal acquisition (1 kHz sampling)

Performance Metrics and Clinical Validation

Quantitative Performance Assessment

Recent studies have demonstrated promising performance for piezoelectric sensor-based dysphagia assessment. A systematic review of AI-based dysphagia screening technologies reported accuracy ranging from 71.2% to 99% and sensitivity ranging from 63.6% to 100% across various sensing modalities [1]. The integration of artificial intelligence, particularly support vector machines (62% of studies) and emerging deep learning approaches, has enhanced detection capabilities [1]. Multimodal sensing approaches generally outperform single-modality systems by capturing complementary aspects of swallowing physiology [1].

Table 3: Performance Metrics of Dysphagia Assessment Technologies

Assessment Method	Key Performance Metrics	Advantages	Limitations
Piezoelectric Sensor Arrays [11] [12]	Laryngeal movement velocity: 0.08-0.11 m/s; Swallowing latency: ~0.5 s	Quantitative kinematics; non-invasive; suitable for repeated measures	Requires skin contact; positioning sensitivity
AI-Enhanced Screening [1]	Accuracy: 71.2-99%; Sensitivity: 63.6-100%; AUC: 0.77-0.977	Objective classification; potential for automation	Model training requirements; computational complexity
Acoustic Methods [2] [13]	High temporal resolution; swallowing sound identification	Non-contact operation; continuous monitoring	Background noise susceptibility; privacy considerations
Clinical Gold Standards [10] [1]	High diagnostic accuracy for aspiration detection	Comprehensive anatomical and functional assessment	Radiation exposure (VFSS); invasiveness (FEES); limited accessibility

Implementation Framework and Future Directions

The transition from laboratory demonstration to clinical implementation requires addressing several key challenges. Sensor adhesion and positioning remain critical for consistent signal acquisition, with advances in biocompatible adhesives and anatomical conformability improving reliability [11] [10]. Signal standardization across individuals with varying neck anatomies necessitates adaptive algorithms and potential multi-sensor arrays to account for anatomical differences [12]. Integration with existing clinical workflows will require user-friendly interfaces and automated interpretation systems to facilitate adoption by healthcare providers with varying technical expertise [10] [1].

Future development should focus on multi-parametric sensing systems that combine piezoelectric sensors with complementary modalities such as accelerometry or acoustics to enhance detection specificity [1]. The creation of lead-free piezoelectric materials with improved performance characteristics will address both environmental concerns and biocompatibility requirements [14] [6]. Longitudinal monitoring capabilities through wearable, wireless systems will enable the capture of spontaneous swallowing patterns in natural environments rather than controlled clinical settings [11] [10]. Finally, the implementation of edge computing and real-time feedback could transform these systems from assessment tools to therapeutic devices that provide biofeedback during swallowing rehabilitation [10].

Piezoelectric sensor technology represents a promising pathway toward addressing the critical clinical need for accessible dysphagia screening. The development of flexible, sensitive, and biocompatible sensors enables non-invasive detection of laryngeal movement with quantitative precision comparable to traditional clinical assessments. By leveraging advancements in materials science, particularly aluminum nitride thin films and lead-free bismuth iodide hybrids, alongside sophisticated signal processing and pattern recognition algorithms, these systems offer the potential to transform dysphagia care from episodic, clinic-based assessment to continuous, natural environment monitoring. Future research should focus on validating these technologies in diverse patient populations and integrating them into comprehensive clinical pathways to improve early detection, rehabilitation, and quality of life for individuals with swallowing disorders.

The accurate detection of laryngeal movement and hyoid bone displacement is paramount for assessing swallowing function, diagnosing dysphagia, and advancing research in drug development for neuromuscular disorders. These two anatomical structures are central to the coordinated biomechanical events that ensure safe and efficient bolus transport from the oral cavity to the esophagus while protecting the airway [10]. The hyoid bone serves as a central attachment point for the suprahyoid and infrahyoid muscles, and its characteristic upward and forward motion is a key kinematic event in swallowing. This movement is mechanically linked to the elevation of the larynx, which facilitates the closure of the laryngeal vestibule and the opening of the upper esophageal sphincter [15] [16].

Non-invasive sensor technologies, particularly piezoelectric sensors, have emerged as powerful tools for quantifying these movements outside restrictive clinical settings. This application note details the experimental protocols and quantitative findings from recent studies utilizing these technologies, providing a framework for researchers and scientists to implement these methods in their own work on chewing and swallowing detection.

The following tables consolidate key quantitative findings from recent studies on sensor-based measurement of laryngeal and hyoid kinematics.

Table 1: Temporal and Velocity Parameters of Laryngeal Movement from a Piezo Sensor Array Study (n=12) [17]

Parameter	Men (Mean)	Women (Mean)
Maximum Rising Velocity	0.08 m/s	0.11 m/s
Maximum Lowering Velocity	0.09 m/s	0.11 m/s
Swallowing Latency	0.49 s	0.53 s

Table 2: Sensor Performance Characteristics for Swallowing Detection

Sensor Type	Target Anatomy/Metric	Key Performance Metric	Value	Source
Piezoelectric (PVDF)	Jaw; Chewing	F1-score for Chewing Detection	0.94	[18]
Accelerometer & Gyroscope	Wrist; Eating Gesture	F1-score for Eating Gesture Detection	0.82	[18]
RIP Sensor*	Chest/Abdomen; Swallowing	F1-score for Swallowing Detection	0.58	[18]
FSR Sensor	Thyroid Cartilage Excursion	Correlation with VFSS (Temporal Parameters)	R = 0.813 - 0.999	[16]
Photoelectric Sensor (Nodomiru)	Larynx; Up-Down Movement	Intrarater Reliability (ICC)*	0.694 - 0.967	[19]

RIP: Respiratory Inductance Plethysmographic. FSR: Force-Sensing Resistor. *ICC: Intraclass Correlation Coefficient, for the average of five measurements.

Experimental Protocols

Protocol 1: Laryngeal Movement Detection with a Flexible Piezoelectric Sensor Array

This protocol is adapted from a study employing a custom-built array of small piezo pressure sensors to non-invasively detect laryngeal movement [17].

3.1.1 Research Reagent Solutions

• Piezoelectric Sensor Array: Five small piezo pressure sensors (1.5 mm length, 7.0 mm width) made from Polyvinylidene fluoride (PVDF) sheets, lined up with 3.0-mm intervals, embedded in a palm-sized urethane resin sheet.
• Signal Conditioning Unit: An impedance conversion circuit with a total gain of 0.56 and a time constant of 3.0 s.
• Data Acquisition System: An analog-digital converter (e.g., CED Power1401-3) and software (e.g., Spike2) with a sampling frequency of 1 kHz.

3.1.2 Procedure

• Subject Preparation: Seat the subject on a backless chair. Instruct the subject to hold a 3 mL bolus of water in their mouth.
• Sensor Placement: Lightly attach the sensor sheet to the ventral surface of the neck near the laryngeal prominence. Position the lowest sensor 0.5–1.0 cm higher than the laryngeal prominence at rest.
• Data Recording: Within 30 seconds of attachment, instruct the subject to swallow and simultaneously push a foot switch to mark the event.
• Data Acquisition: Record signals from all five sensor channels and the foot switch into the PC via the analog-digital converter.
• Repetition: Repeat the procedure for 10-20 swallows per subject.
• Analysis: Identify the first and second peaks in the sensor signal, which correspond to the upper and lower positions of the larynx during swallowing. Calculate velocities and latencies from the signal.

Protocol 2: Validation of Laryngeal Excursion Measurement Using Force-Sensing Resistor (FSR) Sensors

This protocol outlines a method for validating a non-invasive FSR sensor against the gold-standard Videofluoroscopic Swallowing Study (VFSS) [16].

3.2.1 Research Reagent Solutions

• FSR Sensors: Two FSR sensors (e.g., part of an MP150 system, Biopac), one for detecting thyroid cartilage excursion and another for thumb pressing.
• Videofluoroscopy System: A VFSS system (e.g., Siemens Luminos) recording at 30 frames/s.
• Additional Sensors: Submental surface electromyography (sEMG) electrodes and a nasal cannula for respiratory monitoring.
• Analysis Software: AcqKnowledge for data recording and LabView for offline analysis.

3.2.2 Procedure

• Subject Preparation: Recruit healthy volunteers. Place two ECG adhesive electrodes for submental sEMG, a nasal cannula for respiration, and an FSR sensor on the anterior neck at the thyroid cartilage level.
• Reference Sensor: Have the subject hold a second FSR sensor fixed to a small stick in their hand to use as an event marker via thumb pressing.
• Synchronized Recording: In the VFSS suite, administer a 5 mL liquid barium bolus. Instruct the subject in the following sequence: "press" (thumb press), "press," "swallow," "press."
• Simultaneous Data Capture: Simultaneously record the VFSS video and all sensor signals (FSR, sEMG, respiration) from the MP150 system.
• Data Analysis: Export VFSS videos and sensor data. Define key temporal events (onset of hyoid movement, peak hyoid position, etc.) in the VFSS and correlate them with the corresponding points in the FSR sensor signal, using the thumb press events for precise synchronization.

Diagram 1: FSR-VFSS Validation Workflow. This diagram illustrates the experimental protocol for validating Force-Sensing Resistor (FSR) sensor signals against the gold-standard Videofluoroscopic Swallowing Study (VFSS).

The Scientist's Toolkit

Table 3: Essential Research Reagents and Materials for Swallowing Detection Studies

Item	Function/Application	Example Specifications / Notes
Piezoelectric Film (PVDF)	Core sensing element; converts mechanical strain from laryngeal movement into an electrical signal.	PVDF sheets, 40 μm thickness; piezoelectric coefficient d33: 35 pC/N [17].
Flexible AIN Sensor	Highly compliant, thin-film piezoelectric sensor for conformal lamination on neck skin.	Aluminum Nitride on Kapton substrate; ~26 μm thick; integrated with wireless module [11].
Force-Sensing Resistor (FSR)	Detanges changes in force/pressure exerted by thyroid cartilage excursion against the skin.	Used in ambulatory systems; provides good temporal correlation with VFSS [16].
Photoelectric Distance Sensor	Measures distance to skin surface; tracks laryngeal prominence position in anterior-posterior and vertical axes.	Array of 16 sensors at 4mm intervals (e.g., Nodomiru device) [19].
Bend Sensor	Measures the flexion angle of the neck surface during laryngeal elevation.	Thin, flexible membrane; resistance increases with bend angle [15].
Signal Conditioner & DAQ	Amplifies, filters, and digitizes low-voltage analog signals from sensors for PC analysis.	Requires high-input-impedance amplifier for piezoelectric sensors; e.g., Biopac MP150 system [16].
Synchronization Trigger	Enables temporal alignment of data from multiple systems (e.g., sensor data and VFSS video).	A simple FSR used as a hand-operated event marker is effective [16].

Discussion and Technical Considerations

The presented data and protocols demonstrate the viability of piezoelectric and other non-invasive sensors for quantifying laryngeal movement and hyoid bone displacement. Key technical considerations for implementing these methods include:

Sensor Selection and Placement: The choice of sensor depends on the target parameter. Piezoelectric sensors are excellent for capturing the dynamic timing and velocity of laryngeal movement [17]. In contrast, photoelectric sensor arrays can track the trajectory of the laryngeal prominence [19]. Accurate placement over the thyroid cartilage is critical for signal quality and consistency.

Validation Against Gold Standards: For scientific rigor, it is essential to validate non-invasive sensor outputs against established clinical tools like VFSS. The high correlations (R > 0.8) reported between FSR sensor signals and VFSS-measured hyoid motion provide strong support for the validity of these methods for assessing temporal parameters of swallowing [16].

Multi-Sensor Fusion: As shown in Table 2, a single sensor modality may excel at detecting one aspect of swallowing (e.g., chewing) but perform less well on others (e.g., swallowing). Combining sensors that focus on different stages of the dietary cycle—such as an inertial measurement unit for hand-to-mouth gestures, a piezoelectric sensor for chewing, and a respiratory sensor for swallowing—can significantly improve overall eating event detection and resilience against false positives from non-eating activities [18].

Diagram 2: Sensor Types and Their Primary Applications. This diagram classifies common non-invasive sensors used in swallowing research based on their primary measurand and links them to their key applications in detecting laryngeal and hyoid movement.

The instrumental assessment of swallowing function has long been reliant on two gold-standard methodologies: the Videofluoroscopic Swallow Study (VFSS) and the Flexible Endoscopic Evaluation of Swallowing (FEES). While these tools provide critical diagnostic information, they present significant limitations, including patient discomfort, radiation exposure, and restricted accessibility. This has catalyzed the development of non-invasive sensor technologies, particularly those based on piezoelectric principles, which offer a promising alternative for chewing and swallowing detection. This application note details the quantitative advantages of these emerging technologies and provides explicit experimental protocols for their implementation in research settings, framing them within a broader thesis on piezoelectric sensor research.

Table 1: Quantitative Comparison of Swallowing Assessment Technologies

Parameter	VFSS	FEES	Piezoelectric Sensor Array [12]	Neck-worn Electronic Stethoscope [4]	AI-Assisted FEES-CAD [20]
Radiation Exposure	~1.23 mSv (Equivalent to ~10 chest X-rays) [21]	None [22]	None	None	None (inherent to FEES)
Invasiveness	Minimal (contrast agent ingestion)	Moderate (transnasal endoscope insertion) [10]	Non-invasive (skin surface attachment)	Non-invasive (skin surface attachment)	Moderate (inherent to FEES)
Key Measured Variable	Bolus flow, anatomical movement (X-ray)	Pharyngeal and laryngeal anatomy & residues (direct vision) [23]	Laryngeal movement velocity & timing [12]	Swallow sound count & timing [4]	Aspiration/Penetration classification accuracy
Sample Performance Metric	N/A (Gold Standard)	84% Sensitivity, 94% Specificity for aspiration vs. VFSS [24]	Max rising velocity: ~0.08-0.11 m/s; Swallowing latency: ~0.49-0.53 s [12]	Able to measure Oral Processing and Swallowing Time (OPST) objectively [4]	92.5% accuracy for aspiration & penetration detection [20]
Primary Limitation	Radiation exposure, specialized clinic [21]	"White-out" during swallow, patient discomfort [20] [21]	Spatial resolution vs. VFSS/FEES	Requires signal processing for noise filtering	Dependent on quality of FEES procedure

Established Gold Standards: VFSS and FEES

The Videofluoroscopic Swallow Study (VFSS) and Flexible Endoscopic Evaluation of Swallowing (FEES) represent the current clinical benchmarks for diagnosing oropharyngeal dysphagia. A systematic comparison of their diagnostic performance reveals that FEES demonstrates a higher ability to diagnose pharyngeal residue, penetration, and aspiration compared with VFSS, though overall diagnostic performance shows no significant differences, and test choice often depends on availability and patient-specific factors [23]. FEES is particularly noted for its reliability in evaluating patients post-partial laryngectomy, showing good sensitivity (84%) and specificity (94%) for detecting aspiration against the VFSS benchmark [24].

The limitations of these gold standards are well-documented. VFSS involves ionizing radiation with an effective dose of approximately 1.23 mSv per exam, about ten times higher than a standard chest X-ray [21]. This raises concerns for vulnerable populations and limits the feasibility of repeated assessments. FEES, while avoiding radiation, requires transnasal insertion of an endoscope, which can cause patient discomfort and anxiety [10]. Furthermore, a fundamental limitation of FEES is the "white-out" phenomenon, where pharyngeal contraction briefly obstructs the endoscopic view at the precise moment of swallowing, potentially obscuring direct visualization of aspiration [20]. Both procedures require specialized clinical settings and expert interpretation, restricting their accessibility and convenience for routine screening or long-term monitoring [25] [10].

Emerging Non-Invasive Sensor Technologies

Piezoelectric Pressure Sensor Arrays

A significant advancement in non-invasive swallowing detection is the development of flexible piezoelectric pressure sensor arrays. One pioneering device features five small piezo pressure sensors (1.5 mm long, 7.0 mm wide) made from polyvinylidene fluoride (PVDF) sheets, lined up at 3.0-mm intervals and embedded in the middle of a palm-sized, soft urethane resin sheet [12]. This design allows the sheet to be lightly attached to the ventral surface of the neck near the laryngeal prominence to detect movement during swallowing without restricting motion.

The underlying principle involves the piezoelectric effect, where mechanical stress from laryngeal movement generates a measurable electrical voltage. The first and second peaks in the sensor signal correspond to the upward and downward movement of the larynx during a swallow. Research has quantified key swallowing parameters using this technology, including mean maximum rising velocities of 0.08-0.11 m/s and swallowing latencies of approximately 0.49-0.53 seconds in healthy adults [12]. A distinct advantage of this piezoelectric array is its functionality across diverse patient anatomies, including individuals without a prominent larynx, such as many women, where other non-invasive methods like photo-reflective sensors face challenges [12].

Complementary Non-Invasive Modalities

Other non-invasive sensing modalities are emerging alongside piezoelectric sensors, creating a rich ecosystem of alternative technologies:

• Acoustic/Vibratory Sensing: Neck-worn electronic stethoscopes (NWES) utilize piezoelectric vibration sensors to capture swallowing sounds. Coupled with deep learning algorithms, these devices can automatically detect and count swallowing actions during solid food consumption, enabling objective measurement of tests like the Test of Masticating and Swallowing Solids (TOMASS) [4].
• Epidermal Wearable Sensors: The field of skin-interfacing electronics is developing minimally obtrusive sensors for long-term monitoring of swallowing function. These devices aim to measure correlates of swallowing exertions and respiratory activity, offering potential for at-home and continuous monitoring outside clinical settings [10].
• Photoacoustic Imaging: This hybrid modality combines light and sound to visualize swallowing dynamics. A study using a charcoal solution as a contrast agent demonstrated the potential of photoacoustic imaging to detect bolus flow in the airway in real-time, presenting a non-ionizing alternative to VFSS [21].
• AI-Assisted Diagnosis: Artificial intelligence is being integrated with both traditional and novel methods. For instance, FEES-CAD is a convolutional neural network (CNN)-based system that achieves expert-level accuracy (92.5%) in detecting aspiration and penetration from FEES videos, potentially aiding clinicians in interpretation [20].

Experimental Protocols for Sensor Validation

Objective: To validate a flexible piezoelectric pressure sensor array for noninvasive detection of laryngeal movement during swallowing and to quantify key temporal and velocity parameters.

Materials:

• Flexible piezoelectric sensor array (e.g., 5 PVDF sensors embedded in urethane resin sheet)
• Impedance conversion circuit and signal amplifier
• Analog-to-digital converter (e.g., CED Power1401-3)
• Data acquisition software (e.g., Spike2 version 7)
• Web camera for synchronized video recording

Procedure:

• Sensor Preparation: Verify sensor connectivity and calibrate the signal acquisition system. The total gain should be set at 0.56 with a time constant of 3.0 seconds.
• Subject Positioning: Seat the subject on a backless chair to minimize postural adjustments. Record photographs of the front and side views of the neck for anatomical reference.
• Sensor Placement: Lightly attach the sensor sheet to the ventral surface of the neck. Position the lowest sensor 0.5–1.0 cm superior to the laryngeal prominence at rest.
• Task Instruction: Instruct the subject to hold 3 mL of water in their mouth with the tongue tip touching the upper incisors ("tipper" swallow position).
• Data Acquisition: Upon a verbal swallow command, simultaneously trigger data recording and have the subject activate a foot switch. Record signals from all five sensors and the foot switch at 1 kHz.
• Video Recording: Synchronize a web camera (e.g., 320 × 240 pixels at 30 fps) to capture neck movement and environmental sound.
• Data Collection: Repeat steps 4-6 for 10-20 swallows per subject, with adequate rest between trials.

Data Analysis:

• For each swallow, identify the first and second major peaks in the sensor signal, corresponding to superior and inferior laryngeal movement.
• Calculate the maximum rising velocity (peak1 amplitude/rise time) and maximum lowering velocity (peak2 amplitude/fall time).
• Measure swallowing latency as the time interval between the swallow command and the onset of the first sensor signal peak.

Objective: To implement a semi-automated assessment of the Test of Masticating and Swallowing Solids (TOMASS) using a neck-worn electronic stethoscope (NWES) for objective measurement of masticatory and swallowing parameters.

Materials:

• Neck-worn electronic stethoscope (NWES) with piezoelectric vibration sensor
• Smartphone for data recording and video capture
• Commercially available crackers (e.g., 3 g per cracker, Nabisco Premium Crackers)
• Audio-video synchronization software (e.g., ELAN version 6.6)

Procedure:

• Device Setup: Don the NWES around the anterior neck between the C2 and C5 vertebrae. Connect the sensor to a smartphone for data recording.
• Video Setup: Position the smartphone camera to capture a close-up view of the participant's mouth and neck during cracker consumption.
• Test Administration: Provide two separate crackers. Instruct the participant to eat one cracker at a time and verbally indicate "Finished" upon completion.
• Data Recording: Simultaneously record audio data from the NWES and video data from the smartphone throughout the cracker consumption task.
• Data Synchronization: Manually synchronize the audio and video recordings in analysis software by aligning the audio waveform with the video frame of the "Finished" utterance.

Data Analysis:

• Discrete Bite Count: Count the total number of bites from the synchronized video recording.
• Swallow Count: Count the number of swallows using both the video recording and the audio waveform of swallowing sounds.
• Oral Processing and Swallowing Time (OPST): Measure the duration from the initial biting sound to the onset of the "Finished" utterance.
• First Oral Processing and Swallowing Time (1st-OPST): Measure the duration from the initial biting sound to the onset of the first swallow sound.

Research Reagent and Materials Toolkit

Table 2: Essential Research Materials for Non-Invasive Swallowing Detection

Item	Function/Application	Example Specifications/Notes
PVDF Piezoelectric Sensor	Core sensing element for detecting laryngeal movement via pressure changes.	40 μm thickness; theoretical piezoelectric coefficient d33: 35 pC/N; requires stainless steel casing and shield wire [12].
Urethane Resin Substrate	Flexible, skin-compatible base for embedding sensors.	Asker C hardness: 5; thickness: 8 mm; provides comfort and conforms to neck contour [12].
Neck-worn Electronic Stethoscope (NWES)	Piezoelectric vibration sensor for cervical auscultation and swallow sound detection.	Positioned between C2-C5; connects to smartphone for data recording and deep learning-based analysis [4].
Charcoal Contrast Solution	Contrast agent for photoacoustic imaging studies of swallowing.	10 mg/mL concentration in milk/barium base; provides strong photoacoustic signal for bolus tracking [21].
Synchronized Video Recording System	Essential for ground truth validation of sensor data.	High-speed camera (100 fps) recommended; critical for correlating sensor signals with swallowing events [4] [12].
IDDSI Standardized Boluses	Controlled consistency foods and liquids for standardized swallowing challenges.	Range from thin liquids (IDDSI 0) to solids (IDDSI 7); enables systematic assessment across viscosities [24].

Technology Integration and Workflow

The following diagram illustrates the integrated workflow for deploying and validating non-invasive swallowing assessment technologies, from sensor data acquisition to clinical interpretation.

Non-Invasive Swallowing Assessment Workflow

The limitations of traditional gold-standard swallowing assessments have created a compelling need for innovative non-invasive alternatives. Piezoelectric sensor arrays and related technologies represent a promising paradigm shift, offering objective quantification of swallowing parameters without radiation exposure or patient discomfort. The experimental protocols and technical toolkit detailed in this application note provide a foundation for researchers to advance this field, validating these technologies against clinical standards and exploring their potential for accessible, long-term swallowing monitoring in both clinical and community settings.

Sensor Design, Signal Acquisition, and Integrated Systems

The accurate detection of chewing and swallowing is critical for assessing dietary habits, monitoring health conditions, and diagnosing disorders like dysphagia. Piezoelectric sensors, which convert mechanical stress from laryngeal movement into measurable electrical signals, have emerged as a key technology for this purpose due to their sensitivity, non-invasiveness, and versatility [26] [27] [10]. This document details the evolution of sensor architectures from single elements to multi-array configurations, providing application notes and experimental protocols to guide researchers and drug development professionals in implementing these systems for precise biomechanical monitoring.

Sensor Architectures and Performance Analysis

The design of the sensor configuration is a primary factor influencing the type and quality of data obtained. The following section compares different sensor modalities and architectures.

Comparison of Swallowing Detection Modalities

The table below summarizes the key characteristics of different sensing approaches for monitoring swallowing, highlighting the position of piezoelectric sensors within the technological landscape.

Table 1: Comparison of Swallowing Detection and Monitoring Modalities

Sensing Modality	Key Principle	Key Performance Metrics	Advantages	Disadvantages/Limitations
Piezoelectric Sensor Array [26]	Detects mechanical skin movement from laryngeal motion via the direct piezoelectric effect.	Mean maximum rising velocity: ~0.08 m/s (men), ~0.11 m/s (women). Swallowing latency: ~0.49 s (men), ~0.53 s (citation:1).	Non-invasive, can track movement trajectory, suitable for evaluating swallowing function [26].	Physical burden for long-term wear, requires skin contact [13].
Piezoelectric-based Inertial Sensing [28]	Detects vibrations in the neck associated with swallows using a "smart necklace".	F-Measure for food classification: 75.3% - 79.4% [28].	Lower power consumption compared to audio [28].	Lower classification accuracy compared to audio-based methods [28].
Audio-based Detection [28] [13]	Uses a throat microphone to capture swallowing and chewing sounds.	F-Measure for food classification: 88.5% - 91.3% [28].	Higher classification accuracy, comfort (no constant skin contact needed) [28].	Higher computational overhead and power dissipation, privacy concerns [28] [13].
Videofluoroscopy [10]	X-ray video with radiocontrast agent to visualize bolus movement.	Provides real-time visualization of internal swallowing mechanics and aspiration [10].	Clinical gold standard, direct visualization of bolus flow and aspiration [10].	Invasive (radiation exposure), requires clinical setting and experts, not for long-term monitoring [13] [10].
High-Resolution Manometry [10]	Measures internal pharyngeal and esophageal pressures via a transnasal catheter.	Provides quantitative pressure data and coordination timing of the swallow [10].	Highly quantitative pressure data [10].	Minimally invasive, requires clinical expert, obtrusive [10].

Evolution of Piezoelectric Sensor Configurations

Piezoelectric sensing systems for deglutition have evolved in complexity to capture richer data.

• Single-Element Sensors: Early and simpler systems utilize a single piezoelectric sensor placed on the neck. These are effective for basic swallow counting or detecting the presence of a swallow event [28]. However, they offer limited spatial information and cannot characterize the direction or precise pattern of laryngeal movement.

• Multi-Array Architectures: To overcome the limitations of single elements, multi-array configurations have been developed. One prominent example involves lining up five small piezo pressure sensors (1.5 mm length, 7.0 mm width) with 3.0-mm intervals, embedded in a palm-sized urethane resin sheet [26]. This array is attached to the ventral surface of the neck near the laryngeal prominence. During a swallow, the sequential activation of sensors in the array allows the device to capture the upward and downward trajectory of the larynx, providing kinematic data such as velocity [26]. This configuration is particularly useful for evaluating swallowing function by characterizing the movement itself.

The logical relationship between the system components and the data workflow in a typical piezoelectric sensing study can be visualized as follows:

Detailed Experimental Protocols

Protocol: Deployment of a Flexible Piezoelectric Sensor Array for Laryngeal Movement Analysis

This protocol is adapted from the work of Iizuka et al. (2018) for evaluating swallowing function by capturing laryngeal kinematics [26].

1. Objective: To non-invasively detect and characterize laryngeal movement during swallowing using a flexible piezoelectric pressure sensor array, obtaining metrics such as maximum rising/lowering velocity and swallowing latency.

2. Research Reagent Solutions & Materials:

Table 2: Essential Materials for Piezoelectric Swallowing Detection Experiments

Item Name	Function/Description	Specific Example / Properties
Flexible Piezoelectric Sensor Array	Core sensing element; converts laryngeal movement mechanical stress into electrical voltage.	Five sensors (1.5mm x 7.0mm) spaced 3.0mm apart, embedded in urethane resin sheet [26].
Signal Amplifier & Conditioner	Amplifies and filters the raw, low-voltage signal from the piezoelectric elements for accurate measurement.	Typically includes high-impedance amplifiers and band-pass filters to remove noise.
Data Acquisition (DAQ) System	Converts the analog voltage signal from the amplifier into a digital signal for computer analysis.	A system with sufficient sampling rate (e.g., 1 kHz) and resolution to capture swallow dynamics.
Secure Attachment Band	Gently secures the sensor array to the participant's neck without restricting movement.	Soft, adjustable strap (e.g., elastic or Velcro) to ensure consistent sensor-skin contact.
Calibration Fixture	Provides a known mechanical input for calibrating the sensor output voltage.	A device capable of applying controlled, small-scale displacements or vibrations.

3. Methodology:

• Sensor Placement: Lightly attach the sensor array sheet to the ventral surface of the participant's neck, ensuring the line of sensors is centered over the laryngeal prominence (Adam's apple). The sheet should be secured using a gentle, hypoallergenic medical adhesive or a soft strap to maintain consistent contact without causing discomfort or restricting swallowing [26].
• Data Collection: Connect the sensor array to the signal conditioner and DAQ system. Instruct the participant to perform a series of water swallows (e.g., 5-10 repetitions) as naturally as possible. Record the voltage-time data from all sensor channels simultaneously throughout each swallow.
• Signal Processing: Offline, apply a digital band-pass filter (e.g., 0.1 - 10 Hz) to the raw data to remove high-frequency noise and slow baseline drift. For each swallow and each sensor channel, identify the first (upper movement) and second (lower movement) major signal peaks.
• Data Analysis:
  • Velocity Calculation: For each peak, calculate the maximum rising and lowering velocity from the slope of the voltage signal before and after the peak, respectively. Convert these values to physical velocities (m/s) using the sensor's calibration factor [26].
  • Swallowing Latency: Calculate the time difference between the first and second peaks, which corresponds to the period the larynx remains in an elevated position during the swallow [26].
  • Statistical Analysis: Report mean and standard deviation for velocities and latencies across all swallows and participants. Compare results between demographic groups (e.g., men vs. women) using appropriate statistical tests (e.g., t-test).

The workflow for this specific protocol, from preparation to analysis, is outlined below:

Protocol: Comparative Assessment of Piezoelectric vs. Audio-based Swallow Detection

This protocol is based on the comparative study by Alshurafa et al. (2016), focusing on classifying food intake from swallows [28].

1. Objective: To objectively compare the classification accuracy and system power requirements of a piezoelectric-based inertial sensing system versus an audio-based detection system for monitoring dietary intake.

2. Materials: - Piezoelectric System: A piezoelectric sensor embedded in a "smart necklace" form-factor to monitor vibrations in the lower neck [28]. - Audio System: A commercial throat microphone placed loosely in the lower part of the neck [28]. - Synchronized DAQ System: A device capable of recording data from both sensors simultaneously. - Food Items: A variety of test foods with different textures (e.g., sandwich, chips, nuts, water).

3. Methodology: - Participant Recruitment: Recruit a cohort of participants (e.g., n=20) with a diverse age range [28]. - Experimental Procedure: Fit each participant with both the piezoelectric necklace and the throat microphone. Instruct them to consume each test food item in a randomized order. Record data from both systems simultaneously during the entire eating period. - Data Labeling: Accurately label the recorded data with the corresponding food type and swallow events to create a ground-truth dataset. - Feature Extraction & Classification: - For the audio-based approach, use a tool like openSMILE to extract a large set of audio features (MFCC, PLP, spectral features, etc.) from 1-second sample windows. Train a classifier (e.g., Random Forests) to distinguish between food types [28]. - For the piezoelectric-based approach, extract statistical features (e.g., mean, variance, peak counts) from the inertial sensor data in the same time windows and train a similar classifier [28]. - Performance & Power Evaluation: Evaluate both systems using metrics like Precision, Recall, and F-Measure via cross-validation. Separately, model the power consumption of each system based on sample rate, computational overhead, and data transmission requirements [28].

The Scientist's Toolkit

A selection of key reagents and materials critical for research in this field is provided below.

Table 3: Key Research Reagent Solutions for Piezoelectric Swallowing Detection

Reagent/Material	Function in Research	Specific Research Application
Urethane Resin Substrate	Serves as a flexible, biocompatible carrier for sensor elements.	Used to embed and protect a linear array of piezoelectric sensors for placement on the neck [26].
Throat Microphone	Captures acoustic signals from the throat for comparative analysis with vibrational data.	Serves as the audio-based reference in comparative studies classifying food types from swallowing sounds [28].
Hybrid CTC/Attention Model	A machine learning architecture for sequence-to-sequence modeling.	Can be adapted from speech recognition to automatically detect and segment chewing and swallowing events from time-series sensor data, using weakly labeled data [13] [29].
Barium Sulfate Radiocontrast	Makes bolus visible under X-ray imaging.	Used in Videofluoroscopic Swallow Studies (VFSS) to provide a clinical gold standard for validating piezoelectric sensor signals against actual bolus movement [10].

Piezoelectric materials, which convert mechanical stress into electrical signals, form the cornerstone of modern self-powered sensing systems. For research in chewing and swallowing detection, these materials offer the unique potential for creating unobtrusive, continuous monitoring devices that do not require external power sources. The intrinsic electromechanical coupling in these materials enables the direct translation of biomechanical forces—such as jaw movements and laryngeal excursions during swallowing—into quantifiable electrical signals. This application note provides a detailed comparison of polyvinylidene fluoride (PVDF), lead-free ceramics, and their nanocomposites, focusing on their selection, processing, and implementation for specific aspects of deglutition monitoring. The growing regulatory pressure against lead-based materials like PZT (lead zirconate titanate) has accelerated the development of lead-free alternatives, making this analysis particularly timely for researchers developing new biomedical devices [30] [31].

Material Properties and Selection Criteria

Key Piezoelectric Properties for Biomechanical Sensing

The performance of piezoelectric materials in swallowing and chewing detection depends on several key properties. The piezoelectric charge constant (d₃₃) indicates the material's sensitivity to applied mechanical stress, directly affecting signal amplitude from subtle swallowing motions. The voltage constant (g₃₃) relates to output voltage per unit stress, crucial for detecting low-force swallows. Electromechanical coupling factor (k) determines energy conversion efficiency, while dielectric constant (ε) affects electrical impedance matching with measurement circuitry. Mechanical flexibility is paramount for wearable sensors conforming to skin contours, and biocompatibility ensures safety for potential skin contact or implantation [32] [33].

Comparative Analysis of Piezoelectric Materials

Table 1: Quantitative Comparison of Piezoelectric Materials for Biomedical Sensing

Material	d₃₃ (pC/N)	g₃₃ (mV·m/N)	ε	Flexibility	Biocompatibility	Key Advantages
PVDF [33]	20-40	200-300	9-12	Excellent	High	High flexibility, simple processing
PVDF Nanofiber [34]	~35	~300	~10	Exceptional	High	Enhanced β-phase, scalable production
KNN Ceramics [30] [35]	80-450	20-35	500-2000	Poor	Moderate	High d₃₃, environmentally friendly
BNT Ceramics [36]	70-150	15-25	400-800	Poor	Moderate	Good temperature stability
BT Ceramics [30]	150-500	10-25	1500-4000	Poor	Moderate	Highest d₃₃ among lead-free options
PVDF-BTO Nanocomposite	45-75	100-200	15-40	Good	High	Balanced performance, enhanced sensitivity

Table 2: Qualitative Application Suitability for Swallowing Detection

Material	Wearable Swallowing Sensors	Implantable Sensors	High-Frequency Monitoring	Long-Term Stability	Processing Complexity
PVDF	Excellent	Good	Good	Good	Low
PVDF Nanofiber	Excellent	Good	Good	Moderate	Moderate
KNN Ceramics	Poor (rigid)	Fair (with encapsulation)	Excellent	Excellent	High
BNT Ceramics	Poor (rigid)	Fair (with encapsulation)	Excellent	Excellent	High
BT Ceramics	Poor (rigid)	Fair (with encapsulation)	Excellent	Excellent	High
PVDF-BTO Nanocomposite	Good	Good	Good	Good	Moderate

For chewing and swallowing detection, material selection depends on the specific monitoring approach. PVDF and its nanofiber forms offer optimal performance for epidermal wearables due to their inherent flexibility, making them ideal for sensing laryngeal movement when attached to neck skin [10]. Lead-free ceramics like KNN and BNT provide superior piezoelectric coefficients but lack flexibility, potentially limiting their use to rigid substrates or hybrid designs where their enhanced sensitivity can be leveraged without direct skin contact. Nanocomposites strike a balance, offering improved sensitivity over pure polymers while maintaining adequate flexibility for wearable applications [31] [32].

Experimental Protocols for Material Processing and Characterization

Protocol 1: Fabrication of PVDF Nanofibers via Solution Blow Spinning

Purpose: To create flexible PVDF nanofiber mats with enhanced β-phase content for high-sensitivity swallowing sensors.

Materials and Equipment:

• PVDF powder (e.g., Kynar 761)
• N,N-Dimethylformamide (DMF) solvent
• Aluminum electrodes
• Solution blow spinning apparatus with concentric nozzle
• Syringe pump (capable of 7 mL/h)
• Air compressor (0.4-0.6 bar)
• Drum collector (7.62 cm diameter)
• Hot plate with magnetic stirring

Procedure:

• Solution Preparation: Dissolve 12 w.% PVDF in 15 mL DMF solvent. Heat and stir at 75°C for 180 minutes until fully dissolved.
• Cooling: Allow solution to cool for 20-25 minutes at 20±2°C and 65±5% relative humidity.
• SBS Setup: Load 15 mL PVDF solution into 20 mL syringe fitted with 23-gauge needle. Mount syringe in pump set to 7 mL/h feed rate.
• Fiber Production: Initiate air flow at 0.4-0.6 bar pressure. Position collector 25 cm from nozzle. Start drum rotation at 1.4 rpm.
• Collection: Continue process for 55-60 minutes to form uniform nanofiber mat.
• Electrode Integration: Sandwich NF-PVDF mat (1.5 × 4 cm) between aluminum electrodes.

Quality Control:

• Verify β-phase content using FTIR spectroscopy (peaks at 840 and 1275 cm⁻¹)
• Characterize fiber morphology using SEM (target diameter: 50-500 nm)
• Confirm piezoelectric response using quasi-static d₃₃ measurement [34]

Protocol 2: Formulation of PVDF-BaTiO₃ Nanocomposites

Purpose: To create flexible piezoelectric composites with enhanced sensitivity for low-force swallowing detection.

Materials and Equipment:

• PVDF pellets
• Barium titanate (BaTiO₃) nanoparticles (≤100 nm)
• DMF or NMP solvent
• Ultrasonic probe homogenizer
• Vacuum oven
• Hot press
• Poling equipment (high-voltage DC source)

Procedure:

• Solution Preparation: Dissolve 15 w.% PVDF pellets in DMF at 60°C with stirring.
• NP Dispersion: Disperse 5-15 w.% BaTiO₃ nanoparticles in minimal solvent using ultrasonic homogenization (30 minutes, 50% amplitude).
• Mixing: Combine PVDF solution and NP dispersion with vigorous stirring for 60 minutes.
• Film Casting: Pour mixture onto glass plate, doctor blade to 100-200 μm thickness.
• Solvent Removal: Dry at 80°C for 4 hours, then vacuum dry at 60°C for 12 hours.
• Thermal Treatment: Anneal at 140°C for 2 hours to enhance crystallinity.
• Poling: Apply DC field of 50-100 MV/m at 80°C for 30 minutes to align dipoles.

Characterization:

• Measure d₃₃ using quasi-static meter (target: 45-75 pC/N)
• Determine β-phase content via XRD (peak at 20.6°) and FTIR
• Evaluate mechanical properties via tensile testing [31] [33]

Protocol 3: Performance Evaluation for Swallowing Detection

Purpose: To characterize piezoelectric material response under simulated swallowing conditions.

Materials and Equipment:

• Customized PCB drilling machine or mechanical tester
• Keithley SourceMeter 2401 or equivalent
• Force sensor
• Signal conditioning circuitry
• Data acquisition system
• Simulated neck phantom

Procedure:

• Sample Mounting: Secure piezoelectric sensor on simulated neck phantom at laryngeal prominence location.
• Force Calibration: Program mechanical tester to apply cyclic forces (0.2-0.4 N) at physiological swallowing frequency (0.6-1 Hz).
• Electrical Measurement: Connect sensor electrodes to sourcemeter configured for voltage and current measurement.
• Signal Acquisition: Record open-circuit voltage and short-circuit current during force application.
• Data Analysis: Calculate power output (P = V×I), sensitivity (V/N), and signal-to-noise ratio.
• Frequency Response: Sweep frequency from 0.1-10 Hz to determine optimal response range.

Expected Outcomes:

• PVDF nanofibers: ~5-10 V output at 0.4 N force
• Power density: 0.12 µW/mm² for NF-PVDF
• Clear differentiation between swallowing and talking signals [34]

Implementation in Chewing and Swallowing Detection

Sensor Design and Integration Strategies

The complex biomechanics of deglutition require thoughtful sensor design that accounts for anatomical variation and movement dynamics. For laryngeal movement detection during swallowing, PVDF-based sensors offer optimal performance due to their flexibility and medium-range piezoelectric coefficients. Epidermal sensors should be designed with a multilayer structure: piezoelectric active layer, electrode layers (often aluminum or flexible ITO-PET), and protective encapsulation (medical-grade silicone or polyurethane) [10].

For comprehensive swallowing assessment, multiple sensor arrays can capture the spatiotemporal sequence of laryngeal elevation. Sensor placement should target:

• Thyrohyoid region (superior laryngeal movement)
• Cricoid cartilage level (inferior movement)
• Submental region (suprahyoid muscle activity)

Lead-free ceramics like KNN can be incorporated in hybrid designs where their superior sensitivity is leveraged in locations not requiring direct skin contact, such as embedded in clothing collars or wearable housings [35].

Signal Processing and Data Interpretation

Piezoelectric signals from swallowing events require specialized processing to distinguish from artifacts and other neck movements:

• Signal Conditioning: High-impedance buffers for PVDF outputs, appropriate filtering (0.1-10 Hz bandpass for swallowing)
• Feature Extraction: Signal amplitude, duration, rise time, and integrated area
• Classification: Machine learning algorithms (Support Vector Machines commonly used) to distinguish safe swallows, weak swallows, and aspiration events [1]

Table 3: Performance Metrics of Piezoelectric Swallowing Sensors

Parameter	Target Value	Measurement Method	Clinical Significance
Sensitivity	>5 V/N	Calibrated force application	Detection of weak swallows
Frequency Response	0.1-10 Hz	Frequency sweep	Capture of rapid swallowing events
Power Output	>0.1 µW/mm²	Resistive load testing	Self-powered operation capability
Signal-to-Noise Ratio	>20 dB	Comparison with baseline	Reliable event detection
Cycle Lifetime	>10⁶ cycles	Repeated mechanical loading	Long-term monitoring viability
Temperature Stability	±5% (20-40°C)	Environmental chamber	Consistent performance on skin

The Researcher's Toolkit

Table 4: Essential Research Reagents and Equipment

Item	Function	Examples/Specifications
PVDF Polymer	Piezoelectric matrix material	Kynar 761, Sigma-Aldrich 427152
BNT/KNN/BT Powders	Lead-free ceramic fillers	99.9% purity, particle size <100 nm
DMF Solvent	Polymer dissolution	Anhydrous, 99.8% purity
Solution Blow Spinning System	Nanofiber fabrication	Custom apparatus with precise airflow control
Ultrasonic Homogenizer	Nanoparticle dispersion	500W, with probe tip
High-Voltage Poling Source	Dipole alignment	0-30 kV DC, temperature chamber
FTIR Spectrometer	β-phase quantification	Resolution 4 cm⁻¹, ATR accessory
Source Measure Unit	Electrical characterization	Keithley 2401, pA measurement capability
Programmable Mechanical Tester	Simulated swallowing forces	0-10 N range, 0.1-20 Hz frequency

PVDF, lead-free ceramics, and their nanocomposites each offer distinct advantages for chewing and swallowing detection applications. PVDF provides exceptional flexibility and processing ease for wearable sensors, while lead-free ceramics like KNN and BNT offer higher sensitivity for applications where rigidity is acceptable. Nanocomposites represent a promising middle ground, balancing enhanced piezoelectric performance with adequate flexibility.

Future research directions should focus on developing multimodal sensing systems that combine piezoelectric sensors with complementary technologies such as electromyography and impedance sensing. Advancements in machine learning for signal classification will enhance the clinical utility of these systems, potentially enabling real-time aspiration risk assessment. Additionally, the development of biodegradable piezoelectric materials would open new possibilities for temporary implantable swallowing monitors without requiring explanation surgery.

The growing regulatory landscape favoring lead-free materials, coupled with advances in material science and fabrication technologies, positions piezoelectric sensors as increasingly viable tools for objective, continuous assessment of swallowing function in both clinical and natural environments.

The integration of wearable platforms and epidermal electronics represents a transformative advancement in the continuous monitoring of physiological signals. These technologies are characterized by their softness, conformability, and biocompatibility, enabling seamless integration with the human body for high-fidelity, long-term health monitoring [37]. For researchers focused on chewing and swallowing detection, this technological evolution is particularly significant. Traditional methods for assessing swallowing function, such as videofluoroscopy (VFSS), are resource-intensive, require specialized equipment, and are unsuitable for routine screening [38] [1]. The emergence of novel sensor technologies, coupled with artificial intelligence (AI), offers a new paradigm for accessible, objective, and reliable monitoring of ingestive behaviors [1] [39].

Within this context, piezoelectric sensors have garnered considerable interest. These sensors can convert mechanical stress from swallowing acoustics and laryngeal movement into quantifiable electrical signals, offering a non-invasive window into deglutition [38] [39]. When these sensors are engineered into epidermal electronic systems—ultrathin, soft, and lightweight devices that conform to the microscale topography of human skin—they overcome the limitations of conventional rigid devices, such as poor skin conformity and motion artifacts [37] [40]. This synergy between innovative sensing mechanisms and advanced platform integration is paving the way for the next generation of intelligent healthcare tools, capable of providing critical data for geriatric care, dysphagia management, and drug development involving swallowable formulations.

Quantitative Performance Data of Selected Sensing Modalities

The performance of sensor-based systems for physiological monitoring can vary significantly based on the sensing modality, data processing approach, and target application. The following tables summarize key quantitative findings from recent research, providing a basis for comparison and selection in system design.

Table 1: Performance of AI-Based Dysphagia Screening Technologies (Adapted from a Systematic Review [1])

Modality Type	Primary Signal	Common AI Model(s)	Reported Performance Range (Accuracy)	Key Advantages
Unimodal	Acoustic	Support Vector Machine (SVM)	71.2% - 99%	Non-invasive, cost-effective
Unimodal	Vibratory	Support Vector Machine (SVM)	71.2% - 99%	Direct mechanical coupling
Multimodal	Acoustic & Vibratory	SVM, Deep Learning	Higher than unimodal systems	Richer feature set, handles multifaceted nature of dysphagia

Table 2: Material Selection for Epidermal Electronics in Wearable Sensors [37] [40] [39]

Material Class	Example Materials	Key Properties	Typical Sensor Type	Suitability for Swallowing Monitoring
Metallic	Gold (Au), Silver (Ag) Nanowires	Superior electrical conductivity (~107 S m⁻¹), chemical stability	Resistive, Piezoresistive	High (Excellent for signal transmission)
Carbon Nanomaterials	Graphene, CNTs	High flexibility, conductivity, durability	Piezoresistive, Capacitive	High (Conformable, sensitive)
Conductive Polymers	PEDOT:PSS	Good flexibility, tunable conductivity	Piezoresistive, Electrochemical	Medium (Balances performance and processability)
Liquid Metals	Eutectic Gallium-Indium (EGaIn)	Extreme stretchability, self-healing	Resistive, Capacitive	High (Withstands large skin deformations)

Table 3: Quantitative Findings from a Pharyngeal Clearance Time (PCT) Study Using a Neck-Worn Electronic Stethoscope [38]

Parameter	Young Group (20-39 years)	Old Group (≥65 years)	Statistical Significance	Correlation with Other Oral Function Items
Pharyngeal Clearance Time (PCT)	Shorter	Significantly longer (P = 0.032)	P = 0.032 between young and old	No significant correlation found
Correlation with Age	Standardized partial correlation coefficient: 0.165 (P = 0.005)	N/A	P = 0.005	N/A

Detailed Experimental Protocols

Protocol 1: Swallowing Acoustics and Pharyngeal Clearance Time Assessment via a Neck-Worn Sensor

This protocol details the methodology for quantifying swallowing function using a neck-worn electronic stethoscope (NWES), as derived from recent research [38].

1. Objective: To non-invasively measure pharyngeal clearance time (PCT) and assess age-related changes in swallowing function in a healthy adult population.

2. Research Reagent Solutions & Equipment:

• Primary Device: Neck-worn electronic stethoscope (NWES) with a piezoelectric vibration sensor (e.g., GOKURI, PLIMES, Inc.) [38].
• Data Acquisition: Smartphone/Tablet with custom Android/iOS application for real-time data reception via Bluetooth Low Energy.
• Calibration Material: 5 mL of water per swallow, administered via a sterile cup or syringe.
• Supporting Equipment: Chair for seated positioning, computer with data processing software (e.g., Python, R, or custom AI engine like INDRA).

3. Procedure: 1. Participant Preparation: Recruit participants based on inclusion/exclusion criteria (e.g., healthy adults ≥20 years, excluding those with neurological disorders or severe dysphagia). Obtain written informed consent. 2. Sensor Placement: Position the NWES sensor over the anterior neck, centered between the C2 and C5 vertebrae, ensuring firm skin contact. 3. Data Recording Setup: Launch the dedicated application on the smartphone. Ensure a stable Bluetooth connection and initiate a new recording session. 4. Swallowing Task: Instruct the participant to hold 5 mL of water in their mouth and swallow upon a verbal cue. Record the swallowing acoustics. 5. Trial Repetition: Allow a brief rest period (e.g., 30 seconds) and repeat the swallow with another 5 mL of water to acquire a duplicate measurement. 6. Signal Processing: The application processes the audio packets (sampled at 11025 Hz). A segment is initiated when a sample's amplitude exceeds a predefined threshold (e.g., 0.17 normalized relative amplitude). The segment ends after five consecutive packets (≈220 ms) fall below the threshold. 7. PCT Extraction: Segments lasting between 0.22 s and 1.76 s are classified as swallows. The duration of this segment is automatically calculated as the PCT, representing the time from the bolus reaching the pharynx to its passage through the upper esophageal sphincter.

4. Data Analysis:

• Calculate the mean PCT from the duplicate swallows for each participant.
• Use statistical tests like two-way ANOVA to compare PCT across predefined age groups and sexes.
• Perform linear regression analysis with PCT as the dependent variable and age/sex as independent variables.

Protocol 2: Integration of Piezoelectric Epidermal Sensors for Laryngeal Vibration Monitoring

This protocol outlines a framework for deploying flexible piezoelectric sensors, representative of the broader class of epidermal electronics, for swallowing detection.

1. Objective: To capture and classify swallowing events via a skin-conformal piezoelectric sensor that detects laryngeal vibrations.

2. Research Reagent Solutions & Equipment:

• Primary Sensor: Flexible piezoelectric acoustic/vibration sensor (e.g., based on a Triboelectric Nanogenerator - TENG) [39].
• Data Acquisition Unit: A lightweight, flexible data logger or a wireless transmitter module capable of interfacing with the sensor.
• Signal Conditioning Circuit: A custom-built or commercial amplifier and filter circuit for the weak piezoelectric signal.
• Machine Learning Platform: Computer with Python/R environment and libraries (e.g., scikit-learn, TensorFlow) for model development.

3. Procedure: 1. Sensor Fabrication & Preparation: Fabricate the piezoelectric sensor using methods such as electrospinning of piezoelectric polymers (e.g., PVDF) or deposition of active materials on a flexible substrate [37] [39]. 2. System Calibration: Calibrate the sensor's response to known mechanical vibrations or sound pressure levels to establish a baseline sensitivity. 3. Skin Attachment: Affix the sensor to the skin over the thyroid cartilage (Adam's apple) using a medical-grade, skin-friendly adhesive. The sensor's nanomesh or serpentine structure should allow for conformal contact without impeding natural movement [37]. 4. Data Collection: Record the sensor's output signal during periods of quiet breathing, saliva swallowing, and water swallowing (e.g., 5 mL and 10 mL boluses). Synchronize data with a video recording or manual event marking for ground truth labeling. 5. Pre-processing: Apply signal processing techniques such as band-pass filtering (e.g., 1-1000 Hz) to remove noise and motion artifacts. Segment the data into individual swallowing event windows. 6. Feature Extraction: Extract relevant features from each event window, including time-domain features (e.g., peak amplitude, duration) and frequency-domain features (e.g., spectral centroid, band energy). 7. Model Training & Classification: Use machine learning algorithms (e.g., Support Vector Machine) to train a classifier to distinguish swallowing events from other activities and to classify different types of swallows [1] [39].

The Scientist's Toolkit: Essential Research Reagents and Materials

Table 4: Key Reagents and Materials for Wearable Swallowing Detection Research

Item Name	Function/Application	Specific Examples / Properties
Piezoelectric Vibration Sensor	Core sensing element for detecting swallowing sounds and laryngeal movement.	Piezoelectric polymer films (e.g., PVDF); Neck-worn electronic stethoscope [38].
Flexible Substrate	Base material providing stretchability and skin conformability for epidermal electronics.	Polyimide, Polydimethylsiloxane (PDMS), Ecoflex; Ultrathin (<5 µm) and soft [37].
Conductive Nanomaterials	Creating stretchable interconnects and electrodes for signal transmission.	Gold (Au) nanomeshes, Silver (Ag) nanowires; Sheet resistance of 4-60 Ω/sq, stretchability up to 40% [37].
Medical-Grade Skin Adhesive	Securely attaches epidermal devices to skin without causing irritation for long-term wear.	Hydrogel-based adhesives, silicone adhesives; Biocompatible, breathable, water-resistant [40].
Support Vector Machine (SVM)	A robust machine learning algorithm for classifying safe vs. unsafe swallows from sensor data.	Most common model in current AI-based dysphagia screening; provides high accuracy [1].
Data Acquisition (DAQ) System	Conditions (amplifies, filters) and digitizes analog signals from the sensor for analysis.	Wireless, low-power DAQ modules; Bluetooth Low Energy for data transmission to smartphone [38] [39].

Workflow and System Integration Diagrams

The following diagrams illustrate the logical workflow for data acquisition and the system-level integration of components for a wearable swallowing monitoring platform.

Workflow for Swallowing Detection and Analysis

System Architecture of a Wearable Swallowing Monitor

The automatic detection of chewing and swallowing is a critical challenge in biomedical engineering, with applications ranging from the management of obesity and dysphagia to drug development and dietary monitoring. Piezoelectric sensors have emerged as a prominent tool for this purpose, capable of capturing the mechano-acoustic signals—including vibrations, sounds, and pressures—generated during ingestive behavior. The efficacy of the final detection or classification system is profoundly dependent on the signal processing workflow that transforms the raw, noisy sensor data into meaningful, discriminative features. This application note details the end-to-end signal processing methodologies, from data acquisition to feature extraction, framed within the context of chewing and swallowing detection research. It provides structured protocols, comparative data, and visual workflows to serve as a practical guide for researchers and scientists developing such systems.

Data Acquisition and Preprocessing

Sensor Types and Data Acquisition Protocols

The first stage in any signal processing workflow is the acquisition of high-fidelity raw data. In the context of chewing and swallowing, several sensor modalities are employed, often in combination.

Table 1: Common Sensor Modalities for Chewing and Swallowing Detection

Sensor Modality	Measured Phenomenon	Typical Sensor Placement	Key Advantages
Acoustic Microphone [41]	Sound waves from swallowing and chewing	Neck (suprasternal notch), external ear canal	Non-invasive; captures rich spectral information
Accelerometer [42]	Skin surface vibrations and movements	Neck (laryngeal prominence), sternal manubrium	Effective for differentiating swallowing from body movement
Piezoelectric Sensor [43] [44]	Mechanical vibration and pressure	Throat, integrated into wearable collars	High sensitivity to vibrations; can be made flexible and wearable
Bioimpedance (BI) [45]	Change in electrical impedance from laryngeal movement	Throat	Directly measures laryngeal closure during a swallow
Electromyography (EMG) [45]	Muscle electrical activity	Submental muscle group	Direct measure of muscle activation during swallowing

A key challenge during acquisition is managing noise and artifacts. As noted in swallowing detection research, signals are frequently contaminated by motion artifacts, intrinsic speech, head movements, and ambient noise [41]. Therefore, acquisition protocols must be carefully designed. This includes using differential sensor configurations, such as two accelerometers placed on the laryngeal prominence and the sternal manubrium, to subtract common-mode noise like whole-body movement [42]. For piezoelectric sensors, an IEPE (Integrated Electronics Piezo-Electric) interface is often necessary to condition the high-impedance signal and provide a constant current source, minimizing noise over long cable runs [46].

Preprocessing and Signal Conditioning

Raw signals require conditioning to enhance the signal-to-noise ratio (SNR) before further analysis. Common preprocessing steps include:

• Filtering: Band-pass filtering is universally applied. For swallowing sounds, a passband of 20-1000 Hz is typical to capture the relevant acoustical information while suppressing low-frequency drift and high-frequency noise [41]. For vibration signals from accelerometers, the frequency range of interest may extend higher.
• Artifact Rejection: Simple algorithms, such as spectral subtraction, can be used for noise reduction, particularly to suppress steady-state ambient noise [47].
• Segmentation: The continuous signal is divided into short, overlapping time windows or epochs for analysis. Epoch size is a critical parameter; while earlier studies used short epochs (12.5–250 ms), research suggests that using epochs closer to the average swallow duration (~860 ms) can improve detection accuracy by capturing the entire event [41].

Feature Extraction Methodologies

Feature extraction is the process of computing quantitative descriptors from the preprocessed signal segments that characterize the underlying physiological event. The methods can be broadly categorized into time-frequency analysis and feature engineering for traditional machine learning.

Time-Frequency Decomposition Techniques

Swallowing and chewing signals are non-stationary, meaning their frequency content changes over time. Time-frequency decomposition is therefore essential.

Table 2: Comparison of Time-Frequency Feature Extraction Methods

Method	Description	Key Parameters	Application in Ingestive Behavior
Mel-Scale Fourier Spectrum (msFS) [41]	Projects the signal onto a bank of frequency filters based on the human auditory perception (mel scale).	Number of mel filters, epoch size, overlap.	Provides features that are perceptually relevant to swallowing sounds; effective for intra-visit models.
Wavelet Packet Decomposition (WPD) [41]	Generalizes Wavelet Transform, providing a rich library of time-frequency bases adaptable to signal characteristics.	Mother wavelet, level of decomposition, chosen nodes for feature calculation.	Offers high accuracy in separating swallowing sounds from artifacts like respiration and speech.
Mel-Frequency Cepstral Coefficients (MFCCs) [29] [47]	Represents the short-term power spectrum of a sound on a non-linear mel scale, followed by a cosine transform.	Number of cepstral coefficients, frame size, number of mel filters.	Captures timbral and textural aspects of chewing and swallowing sounds; widely used in audio processing.
Spectrograms [47]	A visual representation of the spectrum of frequencies in a signal as they vary with time.	Window function (e.g., Hamming), window length, overlap.	Used as input for deep learning models (e.g., CNNs); provides a 2D time-frequency image.

The workflow below illustrates the logical progression from raw data to a set of extracted features ready for model training.

Feature Engineering for Classification

Following time-frequency decomposition, statistical features are often calculated from the resulting coefficients to create a compact feature vector. Common features include:

• Time-Domain: Mean, standard deviation, root mean square (RMS), zero-crossing rate.
• Frequency-Domain: Spectral centroid, spectral roll-off, spectral bandwidth, spectral entropy [47].
• Time-Frequency Domain: For each node in a WPD or each band in a msFS, the mean, energy, and standard deviation of the coefficients are computed.

In a study using bioimpedance (BI) and EMG, the valley depth and width from the BI signal, combined with the integrated EMG activity, were used as features for a support vector machine (SVM) to distinguish swallows from other movements with a sensitivity of 96.1% and specificity of 97.1% [45].

Experimental Protocols for Algorithm Validation

Protocol for Acoustical Swallowing Detection

This protocol is based on a large-scale study that achieved 84.7% accuracy in swallowing event detection [41].

• Subject Population: Recruit subjects with a range of Body Mass Index (BMI), including obese individuals (BMI >30), to ensure the methodology is broadly applicable.
• Data Collection:
  • Use a stethoscope-type electret microphone placed on the neck.
  • Record data during multiple visits, each comprising resting periods and a meal.
  • Include different food textures (e.g., crispy, soft) and conditions (eating in silence and while conversing) to simulate real-world artifacts.
  • Manually label the start and end of all swallowing events in the recorded data to create a ground truth.
• Signal Processing:
  • Preprocess the signal with a band-pass filter (e.g., 20-1000 Hz).
  • Segment the continuous signal into epochs of approximately 860 ms with 50% overlap.
  • Extract features using either Mel-Scale Fourier Spectrum (msFS) or Wavelet Packet Decomposition (WPD). For WPD, a decomposition level of 4 using a Daubechies mother wavelet is a potential starting point.
  • Calculate the mean and energy of the msFS filter outputs or the WPD nodes to form the feature vector.
• Model Training and Validation:
  • Train a Support Vector Machine (SVM) classifier on the extracted features using the manual labels.
  • Use a subject-independent cross-validation strategy (e.g., leave-one-subject-out) to rigorously evaluate generalization performance.

Protocol for End-to-End Deep Learning

This protocol leverages weak labels and has shown high performance in detecting chewing and swallowing [29].

• Data Collection:
  • Use one or more microphones placed under the ear.
  • Record continuous audio during eating sessions.
  • Annotate the data with only the sequence of events (e.g., left chew, right chew, swallow) and their rough timings (weak labels), avoiding the need for frame-accurate labeling.
• Data Augmentation:
  • Apply N-gram-based data augmentation to the weak-labeled sequences to artificially expand the size and diversity of the training dataset.
• Model Training:
  • Utilize a hybrid CTC/attention model architecture. The Connectionist Temporal Classification (CTC) loss function helps in learning the alignment between the input sequence and the output label sequence, while the attention mechanism allows the model to focus on relevant parts of the signal.
  • Input preprocessed log-Mel spectrograms directly to the model.
  • The model outputs a sequence of event labels directly from the input sequence, eliminating the need for manual feature design.

The Scientist's Toolkit: Research Reagent Solutions

Table 3: Essential Research Materials and Tools for Piezoelectric Sensor Research

Item / Solution	Function / Description	Example Use Case
IEPE Piezoelectric Sensor	A sensor with built-in electronics that converts mechanical vibration to a low-impedance voltage signal. Essential for robust data acquisition.	Measuring skin surface vibrations on the neck during swallowing [46].
IEPE Signal Conditioner	Provides the constant current source required to power IEPE sensors and AC-couples the signal to the data acquisition device.	Interfacing an IEPE accelerometer to a standard data acquisition board with a 1 MΩ BNC input [46].
Data Acquisition (DAQ) System	A device that digitizes analog sensor signals. Requires a 24-bit ADC for high dynamic range and sampling rates >2 kHz to capture relevant frequencies.	Converting the analog output from a piezoelectric throat microphone to a digital signal for processing [48].
Support Vector Machine (SVM)	A powerful classifier effective in high-dimensional spaces. Commonly used for classifying engineered features from swallowing signals.	Differentiating swallowing epochs from non-swallowing epochs using features from WPD and msFS [41] [45].
Hybrid CTC/Attention Model	An end-to-end deep learning model that combines CTC loss with an attention mechanism. Ideal for learning from weakly labeled sequence data.	Automatically detecting sequences of left/right chewing and swallowing from continuous audio without precise timing labels [29].
Wavelet Toolbox (e.g., in MATLAB)	A software library providing algorithms and functions for performing Wavelet Packet Decomposition and related analyses.	Extracting time-frequency features from non-stationary swallowing sound signals for event detection [41].

The journey from raw piezoelectric sensor data to discriminative features is a multi-stage process that dictates the success of chewing and swallowing detection systems. Researchers must make critical choices regarding sensor configuration, noise suppression techniques, and the feature extraction paradigm—whether employing well-established time-frequency analysis with feature engineering or adopting modern end-to-end deep learning. The protocols and methodologies detailed herein provide a foundation for developing robust, accurate, and clinically relevant monitoring tools. As sensing technologies and machine learning algorithms continue to advance, these signal processing workflows will remain the core of innovation in the field of ingestive behavior monitoring.

The automatic detection and classification of chewing and swallowing events are critical for health monitoring, managing dysphagia, and behavioral interventions for obesity. Piezoelectric sensors, which convert mechanical deformations from laryngeal movement and muscle activity into quantifiable electrical signals, have emerged as a prominent tool for this purpose due to their sensitivity, non-invasiveness, and suitability for wearable systems. The true potential of these sensors is unlocked through coupling with artificial intelligence (AI), which transforms raw sensor data into meaningful classifications of ingestive behaviors. This application note details the integration of piezoelectric sensors with machine learning (ML) and deep learning (DL) models for the event classification of chewing and swallowing, providing structured data, experimental protocols, and visualization tools to guide research and development in this field.

The following table summarizes the performance of various AI models as reported in recent studies for classifying chewing and swallowing events using sensor data.

Table 1: Performance of AI Models for Chewing and Swallowing Event Classification

AI Model	Sensor Modality	Classification Task	Key Performance Metrics	Reference
Support Vector Machine (SVM)	Piezoelectric strain (temporalis muscle)	Chewing vs. Non-chewing	F-score: 96.28%, AUC: 0.97	[49]
Support Vector Machine (SVM)	Acoustic & Vibratory signals	Dysphagia Screening	Accuracy: 71.2% - 99%, Sensitivity: 63.6% - 100%	[1]
Linear Support Vector Machine	Piezoelectric strain (temporalis muscle)	Chewing segment classification	High accuracy in lab and free-living conditions	[49]
Hybrid CTC/Attention Model	2ch Microphones (under ear)	Left/Right Chewing, Front Biting, Swallowing	Improved detection using weak labels and context	[13]
Deep Learning (e.g., CNN, Transformer)	Depth video, Acoustic signals	Swallowing task classification, Dysphagia screening	Emerging use, high performance potential (e.g., F1 >0.9, Acc 97.8%)	[1] [50]

Detailed Experimental Protocols

Protocol 1: Chewing Bout Detection and Characterization with Piezoelectric Sensor and SVM

This protocol is adapted from a study that achieved high-accuracy chewing detection in both laboratory and free-living environments [49].

1. Sensor System and Instrumentation:

• Primary Sensor: A piezoelectric film sensor (e.g., LDT0-028K from Measurement Specialties Inc.) is used.
• Placement: The sensor is placed on the skin over the temporalis muscle and secured with medical tape. This muscle shows clear activity during jaw movement.
• Device Integration: The sensor is connected to a data acquisition board integrated into the temple of eyeglasses, making the system wearable and socially acceptable.
• Data Acquisition: Signals from the piezoelectric sensor are sampled at 1000 Hz using a 12-bit analog-to-digital converter (ADC).

2. Data Collection Procedure:

• Participant Preparation: Recruit participants without conditions affecting chewing. Instrument them with the sensor system.
• Experimental Sessions: Conduct two distinct sessions:
  • Controlled Laboratory Session: Participants perform activities including quiet rest, consuming a test meal (e.g., cheese pizza), talking, walking on a treadmill, and eating while walking. This sequence captures chewing amidst speech and motion artifacts.
  • Unrestricted Free-Living Session: Participants wear the sensor and follow their daily routine for several hours, including at least one meal episode.
• Data Annotation: In the lab, activities are annotated in real-time by investigators to create a "gold standard" for model training and validation.

3. Signal Processing and AI Model Implementation:

• Energy-Based Segmentation: Instead of using fixed-duration epochs, the raw sensor signal is processed to identify segments of variable length with high energy, which are candidate chewing bouts.
• Feature Extraction: Compute relevant features (e.g., time-domain, frequency-domain) from these segmented signals.
• Model Training and Validation: Train a Support Vector Machine (SVM) model, typically a linear kernel, to classify segments as "chewing" or "non-chewing."
• Chew Count Estimation: A multivariate regression model is used to estimate the number of chews within segments classified as chewing.
• Validation Method: Use 10-fold leave-one-out cross-validation on a participant level to evaluate model performance robustly.

Protocol 2: Multi-Modal Swallowing Dysfunction Assessment

This protocol outlines a methodology for a comprehensive swallowing assessment using multiple sensors, suitable for dysphagia screening [51].

1. Sensor System and Instrumentation: The system employs a multi-modal approach for an integrated assessment:

• Respiratory Flow: Measured using a nasal cannula-type flow sensor connected to a differential pressure transmitter, sampled at 1 kHz.
• Swallowing Sound and Laryngeal Motion: Captured by a custom-made piezoelectric film sensor (e.g., 10mm x 30mm) placed on the thyroid cartilage.
• Signal Separation: The signal from the piezoelectric sensor is split into:
  • High-frequency component (>100 Hz): Represents swallowing sounds.
  • Low-frequency component (<100 Hz): Represents laryngeal motion.

2. Data Collection and Validation:

• Participant Cohort: Includes both healthy volunteers and patients with diagnosed dysphagia.
• Simultaneous Videofluoroscopy (VF): The gold-standard VF is performed concurrently with sensor data collection to provide validated timings and parameters of swallowing events.
• Test Substances: Participants swallow different test foods with varying consistencies (e.g., water, soft jelly, hard jelly, paste), often mixed with a contrast agent for VF.

3. Data Analysis and Feature Extraction:

• Respiratory Phase Analysis: Algorithms classify respiratory activity into inspiration, expiration, and pause. A pause longer than 0.35s is flagged as a potential deglutition apnea.
• Laryngeal Kinematics:
  • Laryngeal Rising Time (LRT): Time taken for the larynx to reach its highest position.
  • Laryngeal Activation Duration (LAD): Duration from the onset of laryngeal elevation to its return to the resting position.
• Swallowing-Breathing Coordination: The system assesses the phase of respiration in which a swallow occurs (e.g., swallowing immediately after inspiration is a risk factor for aspiration).

AI Workflow for Event Classification from Sensor Data

The following diagram illustrates the generalized workflow for processing data from piezoelectric and other sensors to classify chewing and swallowing events using AI.

The Scientist's Toolkit: Research Reagent Solutions

Table 2: Essential Materials and Reagents for Piezoelectric Sensor and AI-Based Detection Experiments

Item Name	Specification / Example	Primary Function in Experiment
Piezoelectric Film Sensor	LDT0-028K (Measurement Specialties); PVDF sheet (e.g., 40μm thick)	Core sensing element; converts mechanical strain from muscle/joint movement into measurable electrical voltage. [49] [12]
Data Acquisition System	Custom board with 12-bit ADC; Microcontroller (e.g., from CED, National Instruments)	Converts analog sensor signals to digital data; often includes signal conditioning (amplification, filtering). [49] [12]
Reference Gold-Standard Equipment	Videofluoroscopy (VF) system; Fiberoptic Endoscopic Evaluation of Swallowing (FEES)	Provides ground truth data for validating the accuracy of the sensor-based AI system. [1] [51]
Multi-Modal Sensors	Nasal cannula respiratory flow sensor; Acoustic microphones (throat or ear)	Provides complementary data streams (respiratory phase, swallowing sounds) to improve classification robustness. [51] [13]
AI/ML Software Platform	MATLAB; Python with libraries (Scikit-learn, TensorFlow, PyTorch)	Environment for implementing signal processing, feature extraction, and training AI classification models. [49] [13]
Wearable Platform	Eyeglasses frame; Neck collar	Houses the sensor and electronics, enabling non-obtrusive monitoring in free-living conditions. [49]

The synergy between piezoelectric sensor technology and advanced AI models creates a powerful toolkit for the objective, accurate, and automated classification of chewing and swallowing events. As evidenced by the protocols and data, SVM currently offers a robust and well-validated approach, while deep learning and end-to-end models like the hybrid CTC/attention present the cutting edge, especially for handling complex, multi-modal data with weak labels. Future work in this field should focus on improving the real-world applicability of these systems through external validation, domain adaptation techniques, and the development of increasingly sophisticated, multi-modal AI architectures to fully address the multifaceted nature of ingestive behaviors and dysphagia.

Overcoming Technical Hurdles and Enhancing Performance

Mitigating Motion Artifact and Environmental Noise

The accurate detection of chewing and swallowing behaviors using piezoelectric sensors is a critical area of research for applications in nutritional science, drug development, and clinical diagnostics. However, the acquisition of these physiological signals is frequently compromised by motion artifacts and environmental noise. Motion artifacts are signal distortions caused by subject movement, which can obscure underlying physiological data and significantly impair the reliability of signal analysis [52]. This document outlines application notes and experimental protocols for mitigating these challenges, framed within broader thesis research on piezoelectric-based detection of masticatory and swallowing functions.

Quantitative Analysis of Motion Artifact Impact and Mitigation Performance

The following table summarizes key quantitative findings from recent literature on the impact of motion artifacts and the performance of various mitigation strategies in the context of physiological monitoring.

Table 1: Quantitative Data on Motion Artifact Impact and Mitigation Performance

Metric	Reported Value / Finding	Context / Sensor Type	Source
Motion Artifact Detection Accuracy	98.61%	Hybrid model (BiGRU–FCN + multi-scale STD) for BCG signals from piezoelectric sensors [52].
Valid Signal Loss in Non-Motion Intervals	4.61%	Hybrid model for BCG signals; indicates efficiency of artifact removal [52].
Chewing Detection Agreement	Correlation coefficient r(550) = 0.955	Strong correspondence between OCOsense glasses (sensing facial movements) and manual video coding [5].
Chewing Count Difference	No significant difference	Between OCOsense algorithm and manual coding [5].
Eating/Non-Eating Behavior Detection	81% (Eating), 84% (Non-Eating)	Classification accuracy of OCOsense glasses [5].
Sensor Stability (ICC)	Good (ICC > 0.80)	Piezoelectric tonometer and accelerometer for pulse waves [53].
	Moderate (0.46 < ICC < 0.86)	Piezoresistive strain gauge and optical sensors for pulse waves [53].
Harmful Vibration Frequency Range	0.5 to 80 Hz	ISO 2631.1 standard for human health; relevant for noise in wearable sensors [54].

Experimental Protocols for Validation and Mitigation

Protocol for Validating Chewing Detection with Piezoelectric Sensors

This protocol is adapted from research validating the OCOsense glasses, which use sensors to detect facial muscle movements [5].

• Objective: To validate the accuracy of a piezoelectric sensor system in detecting and quantifying chewing behavior against a manually coded video ground truth.
• Materials:
  • OCOsense glasses or similar piezoelectric sensor system configured to detect facial muscle movements.
  • Video recording equipment (e.g., smartphone camera).
  • Data synchronization and annotation software (e.g., ELAN, version 6.6 or higher).
  • Test foods (e.g., bagel and apple, as used in the cited study).
• Procedure:
  • Participant Preparation: Recruit healthy adult participants. Exclude individuals with a history of dysphagia or cognitive impairments.
  • Setup: Fit the participant with the sensor system. Position the video camera to capture a clear view of the participant's face and jaw region.
  • Data Collection: Provide the participant with a test food item. Instruct them to eat normally. Simultaneously record sensor data and video.
  • Data Synchronization: Manually synchronize the sensor data and video recordings in the annotation software using a clear event, such as the participant saying "Finished" [4].
  • Ground Truth Annotation: A human coder, blinded to the sensor output, reviews the video to manually annotate the following parameters for each eating segment:
    • Discrete Bite Count: Total number of bites.
    • Chew Count: Total number of chewing cycles.
    • Oral Processing Time: Duration from the first bite to the completion of swallowing.
  • Algorithm Output: Run the sensor data through the proprietary algorithm to obtain its estimates for the same parameters.
  • Data Analysis:
    • Perform regression analysis (e.g., Pearson's r) between the manual coded counts/times and the algorithm-derived counts/times. A strong correlation (e.g., r > 0.95) indicates high validity [5].
    • Use paired t-tests or non-parametric equivalents to check for significant differences in chew counts between methods.

Protocol for Swallowing Detection Using a Neck-Worn Piezoelectric Sensor

This protocol is based on studies utilizing a neck-worn electronic stethoscope (NWES) with a piezoelectric vibration sensor to detect swallowing sounds [4].

• Objective: To objectively measure swallowing counts and timing during solid food consumption using a piezoelectric contact microphone.
• Materials:
  • Neck-worn electronic stethoscope (NWES) with a piezoelectric vibration sensor.
  • Data recorder (e.g., smartphone).
  • Video recording equipment.
  • Test food (e.g., two standard crackers as used in the Test of Masticating and Swallowing Solids - TOMASS).
• Procedure:
  • Sensor Placement: Don the NWES around the anterior neck, positioning the sensor between the C2 and C5 vertebrae [4].
  • Data Collection: Provide the participant with one cracker. Instruct them to eat it at their normal pace. Record swallowing sounds via the NWES and simultaneously video the process.
  • Data Synchronization: Synchronize audio and video files in annotation software using a clear audio-visual marker.
  • Parameter Measurement:
    • Swallow Count: Manually count the number of distinct swallowing sounds in the audio waveform to establish a ground truth [4].
    • Oral Processing and Swallowing Time (OPST): Measure the duration from the initial sound of biting the cracker to the onset of the verbal indication of completion.
    • First Swallow Time (1st-OPST): Measure the duration from the initial bite sound to the onset of the first swallow sound.
  • Automated Detection: Employ a deep learning-based analysis model to automatically identify and count swallowing events from the raw NWES audio data. The model can be trained on features from the time and frequency domains of the signal.
  • Validation: Compare the automated swallow counts and timings against the manually annotated ground truth to calculate accuracy, precision, and recall.

Protocol for a Hybrid Motion Artifact Detection Model

This protocol details the implementation of a high-performance hybrid model for detecting motion artifacts in piezoelectric signals, as demonstrated in ballistocardiogram (BCG) research [52].

• Objective: To implement a dual-channel approach for robust detection of motion artifacts in long-duration physiological signals from piezoelectric sensors.
• Materials:
  • Recorded piezoelectric signal data (e.g., BCG, chewing/vibration signals) containing annotated motion artifacts.
  • Computing workstation with GPU acceleration (e.g., NVIDIA GeForce RTX A2000 or similar).
  • Deep learning framework (e.g., PyTorch in Python).
• Procedure:
  • Data Pre-processing: Segment the continuous signal into windows for analysis.
  • Channel 1: Multi-scale Standard Deviation (STD) Thresholding:
    • Calculate the standard deviation of the signal across multiple window scales.
    • Set empirical thresholds for each scale. A signal segment is flagged as a potential motion artifact if its STD exceeds the threshold at any scale.
  • Channel 2: Deep Learning Classification (BiGRU-FCN):
    • Model Architecture: Construct a hybrid deep learning model comprising:
      • A Bidirectional Gated Recurrent Unit (BiGRU) to capture temporal dependencies in the signal.
      • A Fully Convolutional Network (FCN) to extract hierarchical features.
    • Training: Train the model on labeled datasets where motion artifacts have been manually identified. The input is the raw signal window, and the output is a binary classification (artifact or clean).
  • Hybrid Decision Fusion: A segment is finally classified as containing a motion artifact if it is flagged by either the multi-scale STD thresholding channel or the deep learning channel. This logical OR operation ensures high detection sensitivity.
  • Performance Evaluation:
    • Detection Rate ((R{chk})): Calculate as (R{chk} = N{chk} / N{all}), where (N{chk}) is the number of correctly detected artifact segments, and (N{all}) is the total number of true artifact segments [52].
    • Valid Signal Loss Ratio ((L{effect})): Calculate as (L{effect} = (t{chk} - t{chk_lb}) / (t{all} - t{lb})), where (t{chk}) is the total duration of all detected artifacts, (t{chk_lb}) is the total duration of correctly detected true artifacts, and (t{all} - t{lb}) is the total duration of clean signal in the recording [52]. A low value is desirable.

Signaling Pathways and Workflow Visualizations

Hybrid Motion Artifact Detection Workflow

Experimental Validation Setup for Swallowing

Research Reagent and Essential Materials Toolkit

Table 2: Essential Research Materials for Piezoelectric Chewing/Swallowing Detection

Item / Solution	Function / Application	Example / Specification
Piezoelectric Sensor	Core sensing element converting mechanical strain (vibrations) from chewing/swallowing into electrical signals.	Polyvinylidene fluoride (PVDF) film; Neck-worn electronic stethoscope (NWES) [4]; OCOsense glasses for facial movements [5].
Data Acquisition System	Conditions (amplifies, filters) and digitizes the analog signal from the sensor for processing.	Multichannel physiological recorder (e.g., BL-420S) [53]; Smartphone with audio recording capabilities [4].
Synchronization & Annotation Software	Allows precise temporal alignment of sensor data with video ground truth and manual annotation of events.	ELAN software (version 6.6 or newer) [4] [5].
Test Foods	Standardized stimuli to elicit consistent chewing and swallowing responses for validation studies.	Crackers (e.g., for TOMASS) [4], Bagel, Apple [5].
Deep Learning Framework	Platform for building, training, and deploying models for motion artifact detection and event classification.	PyTorch in Python [52].
High-Performance Computing Workstation	Accelerates the training and evaluation of complex deep learning models on large datasets.	Workstation with GPU (e.g., NVIDIA GeForce RTX A2000) [52].

Ensuring User Comfort and Adherence with Flexible, Biocompatible Designs

The development of flexible, biocompatible sensors represents a paradigm shift in the assessment of chewing and swallowing function. Traditional diagnostic tools for dysphagia, such as videofluoroscopic swallowing study (VFSS) and fiberoptic endoscopic evaluation of swallowing (FEES), present significant limitations including patient discomfort, radiation exposure, and confinement to clinical settings [1] [10]. These factors substantially limit their utility for long-term monitoring and adversely affect patient adherence. Flexible piezoelectric sensors offer a transformative alternative by enabling non-invasive, continuous assessment of swallowing function through the integration of skin-interfacing electronic devices that prioritize user comfort and biocompatibility [10]. The global impact of dysphagia is substantial, affecting 2% to 20% of the general population and up to 80% of older adults with Alzheimer's disease, creating an urgent need for improved monitoring solutions that patients will consistently use [55].

The critical challenge in wearable sensor technology lies in balancing technical performance with user-centered design. Flexible piezoelectric materials (FPM) have emerged as a promising solution, capable of converting mechanical energy from swallowing and mastication into quantifiable electrical signals while maintaining conformity to dynamic human movements [56]. This application note provides a comprehensive framework for the implementation of these sensors, detailing material selection, experimental protocols, and performance metrics essential for ensuring user comfort and adherence in chewing and swallowing detection research.

Performance Characteristics of Detection Modalities

Table 1: Performance Comparison of Swallowing Detection Technologies

Technology Type	Modality	Reported Accuracy	Key Advantages	Comfort Considerations
AI-Based Acoustic Screening [1]	Acoustic Signal	71.2% - 99%	Non-contact; High sensitivity	Minimal physical intrusion; Potential for external placement
Neck-Worn Electronic Stethoscope (NWES) [4]	Cervical Auscultation (Vibratory)	Strong agreement with manual coding	Objective measurement; Semi-automated analysis	Neck-worn; Potential for discreet wear
Piezoelectric Polymers (PVDF) [57] [56]	Mechanical Deformation	14-45.6 V output (cast film)	Excellent flexibility; Biocompatibility	Skin-conformable; Lightweight
Epidermal Sensors [10]	Multiple (EMG, Pressure, Strain)	Varies by signal type	Direct skin interface; Minimal obtrusion	High conformability; Low profile

Table 2: Flexible Piezoelectric Material Properties for Wearable Sensors

Material Category	Representative Materials	Elastic Modulus	Piezoelectric Coefficient (d33)	Comfort & Flexibility Features
Piezoelectric Polymers [57] [56]	PVDF, PAN, PLA, PHB	500-27,900 MPa (PVDF)	-34 pC/N (PVDF, β-phase)	Excellent inherent flexibility; Biocompatible; Processable into fibers
Polymer Composites [56]	PZT/Polymer, ZnO/PVDF	Varies with composition	Significantly higher than polymers	Tailorable flexibility; Balance of performance and comfort
Inorganic Flexible Films [56]	PZT on flexible substrates	High (material-dependent)	High (retains bulk material properties)	Flexibility through thin-film design; May require strategic placement

Experimental Protocol: Assessment of Swallowing Function with Flexible Sensors

Sensor Fabrication and Preparation

Materials Required:

• Polyvinylidene fluoride (PVDF) pellets or pre-fabricated piezoelectric film
• Flexible electrode material (e.g., silver nanowire, conductive carbon ink)
• Biocompatible encapsulation material (e.g., polydimethylsiloxane/PDMS, silicone)
• Laser cutter or precision blade for shaping
• Signal conditioning circuitry (amplification, filtering)

Fabrication Procedure:

• Material Processing: For PVDF-based sensors, process raw material to enhance β-phase content through poling procedures (electrical, thermal, or mechanical stretching) to optimize piezoelectric output [56].
• Electrode Patterning: Deposit flexible electrodes using techniques such as screen printing, vacuum deposition, or solution-based methods, ensuring conformal contact with the piezoelectric layer.
• Encapsulation: Apply a biocompatible encapsulation layer of 100-500μm thickness to protect the sensor from moisture and mechanical damage while maintaining flexibility.
• Signal Conditioning: Integrate flexible circuitry for signal amplification and noise reduction, prioritizing minimal rigidity and maximal comfort.

Participant Preparation and Sensor Placement

Materials Required:

• Medical-grade skin adhesive or hypoallergenic double-sided tape
• Skin preparation supplies (alcohol wipes, mild abrasive)
• Reference sensors for validation (where applicable)

Placement Protocol:

• Site Identification: Identify optimal sensor placement locations based on target physiological signals:
  • Submental Region: For suprahyoid muscle activity during swallowing [10]
  • Anterior Neck (C2-C5): For laryngeal movement detection [4]
  • Temporomandibular Joint Region: For mastication monitoring
• Skin Preparation: Gently clean identified areas with alcohol wipes to remove oils and debris. If necessary, use mild abrasive to reduce skin impedance for EMG applications.
• Sensor Application: Apply sensors using medical-grade adhesive, ensuring secure contact without constricting movement or causing discomfort.
• Comfort Assessment: Query participant regarding comfort perception using a standardized scale (e.g., 1-10 rating) immediately after application and at protocol intervals.

Data Acquisition and Processing

Signal Acquisition Parameters:

• Sampling rate: 1000-4000 Hz for swallowing vibrations [4]
• Gain settings: Appropriate for expected signal amplitude (typically 100-1000x)
• Filtering: Bandpass filter 50-1000 Hz to remove movement artifacts and high-frequency noise

Data Processing Workflow:

• Signal Segmentation: Isolate individual swallow and chew events using amplitude-threshold detection or machine learning approaches.
• Feature Extraction: Calculate relevant temporal and spectral features including duration, amplitude, frequency content, and coordination patterns.
• Validation: Compare sensor outputs with simultaneous video recording or manual annotation to establish accuracy metrics [4] [5].

The Researcher's Toolkit: Essential Materials and Reagents

Table 3: Research Reagent Solutions for Flexible Sensor Development

Category	Specific Materials	Function/Application	Key Considerations
Piezoelectric Polymers [57] [56]	PVDF, PAN, PLA, PHB	Active sensing element; Converts mechanical deformation to electrical signal	β-phase content critical for PVDF performance; Biodegradability of PLA/PHB
Flexible Substrates [56]	PDMS, Ecoflex, Polyimide	Provides structural support; Enables flexibility	Modulus matching to skin (~18.8 MPa) improves comfort
Conductive Elements [10]	Silver nanowires, Graphene, Carbon nanotubes	Electrodes; Signal transmission	Maintain conductivity under strain; Biocompatibility
Encapsulation Materials [10]	Medical-grade silicone, Polyurethane	Protection from environment; Biocompatibility	Moisture barrier properties; Gas permeability for skin health
Adhesives [10]	Medical-grade acrylic, Hydrogel	Secure sensor-skin interface	Hypoallergenic composition; Controlled adhesion strength

Material Selection Framework for Optimal Comfort and Performance

The selection of appropriate materials represents a critical balance between electromechanical performance and user comfort. As illustrated in the decision framework below, researchers must consider multiple factors to optimize adherence without compromising data quality.

Adherence Optimization Strategies

Table 4: Comfort and Adherence Enhancement Techniques

Design Consideration	Implementation Strategy	Expected Impact on Adherence
Mechanical Properties	Match elastic modulus to human skin (~18.8 MPa) [57]	Reduced skin irritation; Improved conformity to movement
Form Factor	Minimal thickness; Anatomical contouring	Discreet wear; Unobstructed daily activities
Breathability	Microperforations; Mesh electrodes	Reduced moisture buildup; Enhanced long-term wear comfort
Adhesive Selection	Controlled adhesion strength; Hypoallergenic composition	Pain-free removal; Minimal skin reactions
Aesthetic Integration	Skin-toned materials; Minimalist design	Reduced stigma; Improved social comfort

Long-term user adherence depends critically on the mitigation of discomfort and integration into daily routines. Research indicates that flexible piezoelectric devices must maintain functionality under repeated bending and stretching conditions while causing minimal interference with normal activities [56]. Furthermore, breathability and skin compatibility become increasingly important for monitoring periods exceeding 24 hours. Implementation of these design principles requires rigorous validation of both technical performance and user experience through structured assessment protocols.

The successful implementation of flexible, biocompatible sensors for chewing and swallowing detection necessitates a multidisciplinary approach that prioritizes user comfort without compromising data integrity. As detailed in these application notes, researchers must carefully consider material properties, sensor design, and implementation protocols to maximize adherence. Future developments in this field will likely focus on further miniaturization, enhanced signal processing algorithms, and more sophisticated biocompatible materials to expand the capabilities of these monitoring platforms. Through adherence to these guidelines, researchers can contribute to the advancement of swallowing science while ensuring that technological solutions remain practical and patient-centered.

Optimizing Sensor Placement for Diverse Anatomies

The accurate detection of chewing and swallowing is critical across numerous fields, including drug development, clinical dysphagia management, and nutritional science. Piezoelectric sensors, which convert mechanical deformations into measurable electrical signals, are particularly well-suited for monitoring the muscular and vibratory activity associated with mastication and deglutition. However, the anatomical variability in craniofacial structure, muscle mass, and subcutaneous tissue between individuals presents a significant challenge for obtaining consistent and reliable signal acquisition. This application note provides detailed protocols for optimizing piezoelectric sensor placement on diverse anatomies, grounded in the physiological mechanisms of eating and the latest advancements in wearable sensor technology. Adherence to these guidelines will enhance data quality, improve detection algorithm performance, and bolster the validity of research outcomes.

Physiological and Anatomical Basis for Sensor Placement

Swallowing (deglutition) is a complex neuromuscular process involving more than 30 pairs of muscles to propel a bolus from the oral cavity to the stomach [10]. It is typically modeled in four stages: oral preparatory, oral propulsive, pharyngeal, and esophageal. The pharyngeal stage is especially critical for dysphagia detection, as it involves the rapid, coordinated action of suprahyoid and thyrohyoid muscles to elevate the hyoid bone and larynx, tilting the epiglottis to seal the airway [10]. These movements create distinct mechanical vibrations and surface deformations that are detectable on the skin.

Similarly, chewing (mastication) involves rhythmic mandibular movement powered by the masseter, temporalis, and other muscles. Research indicates that chewing activates extensive neural networks, including the sensorimotor cortex, insula, cerebellum, and thalamus, and can increase prefrontal cortex activity, reflecting its influence on higher-order cognitive functions [58]. The resulting mandibular movements and muscle contractions provide a robust signal source for piezoelectric sensors. Optimal sensor placement targets superficial anatomical landmarks where these internal activities manifest as measurable surface phenomena.

Piezoelectric Sensor Principles and Performance

Piezoelectric materials, such as lead zirconate titanate (PZT) ceramics or polyvinyl difluoride (PVDF) films, generate an electric charge in response to applied mechanical stress. This direct piezoelectric effect makes them ideal for capturing jaw movements, laryngeal excursions, and skin surface vibrations associated with chewing and swallowing.

A quantitative comparison of sensor modalities highlights that piezoelectric sensors demonstrate good stability (Intra-class Correlation Coefficient, ICC > 0.80) and reproducibility (ICC > 0.75) in physiological signal acquisition [53]. Their performance is, however, dependent on several factors, including the contact force between the sensor and the skin. Studies show that a medium contact force often yields a higher signal-to-noise ratio compared to high or low force levels [53].

Table 1: Quantitative Performance Comparison of Sensor Types for Physiological Monitoring

Sensor Type	Key Measurand	Stability (ICC)	Reproducibility (ICC)	Susceptibility to External Factors
Piezoelectric Tonometer	Pressure / Force	Good (> 0.80)	Good (> 0.75)	Highly dependent on contact force [53]
Accelerometer	Vibration / Acceleration	Good (> 0.80)	Good (> 0.85)	Dependent on contact force [53]
Piezoresistive Strain Gauge (PESG)	Strain / Pressure	Moderate (0.46 - 0.86)	Moderate (0.42 - 0.91)	Dependent on contact force [53]
Optical (PPG)	Blood Volume	Moderate (0.46 - 0.80)	Moderate (0.52 - 0.96)	Susceptible to ambient light [53]

The global piezoelectric devices market, valued at USD 38.40 billion in 2025, is a testament to the growing adoption of this technology across sectors, including biomedical applications [59]. Innovations in flexible and biocompatible piezoelectric materials are further enabling the development of epidermal electronics that conform intimately to the curvilinear surfaces of the human body, improving signal fidelity [10] [60].

Experimental Protocols for Sensor Placement Optimization

Protocol 4.1: Anatomical Landmark Identification and Skin Preparation

Objective: To consistently identify and prepare optimal sites on the skin for piezoelectric sensor attachment across a diverse participant cohort.

Materials:

• Anatomical marker pen
• Disposable alcohol swabs
• Measuring tape or digital calipers
• Skin-safe adhesive (e.g., double-sided dermatological tape)

Procedure:

• Participant Positioning: Seat the participant in an upright position with their head in a neutral posture, looking straight ahead. Ensure the neck muscles are relaxed.
• Landmark Palpation and Marking:
  • Submental Region (Site A): Palpate the area beneath the chin, posterior to the mandibular symphysis. Locate the mylohyoid and digastric muscles by having the participant swallow. Mark a point in the midline of this muscular complex [10].
  • Thyrohyoid Region (Site B): Identify the thyroid cartilage (Adam's apple) and the hyoid bone superior to it. The hyoid bone can be located by having the participant swallow; it is the first structure to move upward. Mark a point on the skin midway between the hyoid bone and the superior border of the thyroid cartilage. This site captures laryngeal elevation [10] [1].
  • Masseteric Region (Site C): Locate the masseter muscle by asking the participant to clench their teeth. Mark a point on the skin over the bulk of the masseter muscle, approximately midway between the angle of the mandible and the zygomatic arch [5].
• Skin Preparation: Gently clean all marked sites with an alcohol swab to remove oils and debris. Allow the skin to air dry completely to ensure optimal adhesion.

Protocol 4.2: Simultaneous Multi-Sensor Signal Acquisition and Validation

Objective: To acquire synchronized data from multiple piezoelectric sensor placements and validate the signals against a ground truth method.

Materials:

• Multiple flexible piezoelectric sensor elements (e.g., PVDF-based)
• Multi-channel data acquisition (DAQ) system
• Synchronized video recording system (e.g., high-frame-rate camera) or surface electromyography (sEMG) system for validation
• Data analysis software (e.g., MATLAB, Python with SciPy)

Procedure:

• Sensor Attachment: Adhere piezoelectric sensors to the prepared Sites A, B, and C. Ensure the sensor axis sensitive to strain or vibration is aligned with the anticipated direction of primary movement (e.g., vertical for laryngeal elevation at Site B).
• System Synchronization: Synchronize the DAQ system's internal clock with the validation system (video/sEMG) using a shared trigger signal or a post-hoc synchronization protocol.
• Data Collection:
  • Instruct the participant to perform a series of tasks in a predetermined order:
    • Resting baseline: 30 seconds of quiet breathing.
    • Water swallows: Three discrete 5 mL water swallows from a cup.
    • Soft food chewing: Chew a 3g piece of bagel for 20 seconds (based on standardized food tests [5]).
    • Hard food chewing: Chew a 3g piece of apple for 20 seconds [5].
  • Record data from all sensors and the validation system simultaneously throughout the protocol.
• Signal Validation: Manually annotate the onset and duration of each swallow and chewing bout from the video recording. For sEMG, use the onset of muscular activity as the gold standard. Cross-correlate these events with the features (e.g., signal peaks, energy) in the piezoelectric sensor data to determine detection accuracy and latency.

Diagram 1: Experimental workflow for multi-sensor signal acquisition and validation.

Protocol 4.3: Signal Processing and Feature Extraction for Detection

Objective: To process raw piezoelectric signals and extract discriminative features for classifying chewing and swallowing events.

Materials:

• Computer with signal processing software
• Custom scripts for feature extraction and analysis

Procedure:

• Pre-processing:
  • Apply a bandpass filter (e.g., 0.5 Hz to 50 Hz) to remove baseline wander and high-frequency noise.
  • Normalize the signal amplitude for each participant to account for inter-subject variability in anatomy and sensor coupling.
• Feature Extraction: For each candidate event, extract the following features from the time-domain signal:
  • Root Mean Square (RMS) Amplitude
  • Signal Entropy
  • Peak Frequency from a Fourier transform
  • Zero-Crossing Rate
• Event Classification: Use a machine learning classifier (e.g., Support Vector Machine - the most common model in recent dysphagia screening studies [1]) trained on the extracted features to automatically classify signal segments as "chew," "swallow," or "noise."

The Scientist's Toolkit: Research Reagent Solutions

Table 2: Essential Materials for Piezoelectric Sensor-Based Chewing/Swallowing Research

Item Name	Function / Application	Technical Notes
Flexible PVDF Piezoelectric Film	Core sensing element; detects skin surface vibrations and deformations from muscle activity.	Biocompatible, flexible, and available in thin sheets. Prefer low-cost variants with good sensitivity for research prototyping [59].
Multi-Channel Data Acquisition (DAQ) System	Amplifies, filters, and digitizes analog signals from multiple piezoelectric sensors.	Requires sampling rate ≥ 1 kHz. Systems with built-in signal conditioning (e.g., National Instruments) are ideal [53].
Dermatological Adhesive Tape	Secures sensors to the skin with minimal irritation for medium-duration studies.	Ensures consistent sensor-skin contact force, a critical factor for signal quality [10] [53].
High-Speed Video Camera	Provides ground truth validation for swallowing and chewing event timing.	Enables manual annotation of swallow onset and chewing bouts, crucial for algorithm training [5].
Standardized Food Boluses	Creates consistent and reproducible stimuli for chewing and swallowing tasks.	Use precisely measured volumes (e.g., 5mL water) and masses (e.g., 3g bagel/apple) to control for bolus effects [5].

Decision Framework for Placement on Diverse Anatomies

Anatomical diversity, such as variations in neck circumference, submental fat, or muscle definition, necessitates an adaptive approach. The following decision framework guides optimal placement.

Diagram 2: Decision pathway for optimizing sensor placement based on individual anatomy.

Guidance Notes:

• For Thin or Muscular Anatomies: The laryngeal prominence and suprahyoid muscles are often easily palpable. The thyrohyoid region (Site B) is typically the optimal location for swallowing detection, as it provides a clear, high-amplitude signal from laryngeal elevation [10].
• For Thick or Obese Anatomies: Subcutaneous tissue can dampen vibrations from the larynx. In these cases, the submental region (Site A) is often more reliable, as the suprahyoid muscles are directly engaged during swallowing initiation and their activity is less dampened. Applying the sensor with a slightly firmer, consistent contact force can improve signal coupling [53].
• Universal Consideration: The masseteric region (Site C) is generally robust for chewing detection across most anatomies, though the precise location along the muscle belly may need minor adjustment via palpation during jaw clenching.

Optimizing piezoelectric sensor placement is not a one-size-fits-all endeavor but a systematic process that must account for anatomical diversity. By adhering to the detailed protocols for landmark identification, multi-sensor validation, and signal processing outlined in this document, researchers can significantly improve the accuracy and reliability of their chewing and swallowing detection systems. The provided decision framework empowers scientists to adapt these protocols effectively across a broad participant spectrum. As the field of epidermal electronics advances, these foundational practices will be crucial for transitioning laboratory proof-of-concepts into validated tools for clinical assessment, drug efficacy testing, and long-term health monitoring.

Power Management and Strategies for Energy Harvesting

The development of automated, wearable systems for detecting chewing and swallowing represents a significant advancement in healthcare monitoring [4] [2] [13]. A critical challenge in deploying these sensor-based systems for continuous, long-term monitoring lies in ensuring a sustainable and reliable power supply without frequent human intervention. Energy harvesting—the technology of deriving power from ambient sources—coupled with sophisticated power management strategies, provides a viable pathway to power these autonomous devices. This application note details the protocols and methodologies for designing efficient energy harvesting and power management systems specifically for wearable sensors used in monitoring masticatory and swallowing functions, with a focus on piezoelectric sensors.

Performance Analysis of Energy Harvesting Techniques

The efficiency of an energy harvesting system is paramount, as it directly impacts the operational uptime and reliability of the wearable sensor. The choice of harvesting technology should be guided by the specific application environment. Table 1 provides a quantitative comparison of the energy conversion efficiencies for various ambient energy sources, contrasting traditional methods with modern Power Management Integrated Circuit (PMIC)-based approaches [61].

Table 1: Energy Conversion Efficiency of Harvesting Techniques

Energy Source	Traditional Harvesting Efficiency (%)	PMIC-Based Harvesting Efficiency (%)
Thermal	5 - 10	70 - 75
Vibrational	10 - 15	80 - 85
Solar	15 - 20	85 - 90
RF	20 - 25	90 - 95

For wearable chewing and swallowing monitors, vibrational energy harvesting is particularly relevant. The mechanical movements of the jaw, neck, and larynx during mastication and deglutition present a consistent energy source that can be scavenged using piezoelectric materials. As shown in Table 1, a PMIC-based system can achieve 80-85% efficiency in converting this mechanical motion into usable electrical energy, a substantial improvement over traditional methods.

The core of an efficient harvesting system is the Energy Harvesting PMIC. This specialized integrated circuit acts as a central power management unit, optimizing the energy transfer from the transducer (e.g., a piezoelectric element) to the electrical load (the sensor and electronics) [61]. Its key functions include:

• Maximum Power Point Tracking (MPPT): Continuously adjusts the electrical operating point of the harvester to draw the maximum possible power, especially crucial under varying physiological activity levels [62].
• Voltage Regulation: Converts the variable voltage from the harvester into a stable, clean supply required by sensitive electronic components.
• Energy Storage Management: Controls the charging and discharging of storage elements to ensure system longevity.

Hybrid Electrical Energy Storage (HEES) Systems

Given the discontinuous nature of both ambient energy and the power consumption of swallowing detection sensors, a robust energy storage solution is essential. A Hybrid Electrical Energy Storage (HEES) system, which combines batteries and supercapacitors, offers an optimal solution [62].

Table 2: Hybrid Electrical Energy Storage (HEES) Components

Storage Device	Function	Key Characteristics
Lithium-ion (Li-ion) Battery	Primary energy buffer for long-term, steady power delivery.	High energy density, moderate self-discharge, requires complex control circuitry [62].
Supercapacitor	Handles high-current, short-duration power pulses (e.g., during sensor activation or data transmission).	High charging/discharging rate, buffers peak loads to protect the battery, extends battery cycle life [62].

The operational strategy for a HEES system is two-fold:

• During Abundant Harvesting: The primary goal is to store as much energy as possible. The PMIC directs harvested energy to power the sensor system first, with excess energy charging the supercapacitor and then the battery. Techniques like task scheduling and dynamic voltage and frequency scaling can be employed to modulate the load's power consumption, thereby increasing the net energy stored [62].
• During Scarce Harvesting: When ambient energy is insufficient, the system draws power from the storage elements. The supercapacitor supplies immediate high-current pulses, while the battery provides the baseline power. Task scheduling is again used to minimize the load's power requirements, thereby extending the system's operational life [62].

Experimental Protocol for System Characterization

This protocol provides a detailed methodology for characterizing the performance of a vibration-based energy harvesting system powering a piezoelectric swallowing sensor.

4.1 Objective To evaluate the energy efficiency, storage capability, and operational stability of a hybrid energy harvesting system under simulated chewing and swallowing activities.

4.2 Materials and Reagents Table 3: Research Reagent Solutions and Essential Materials

Item	Function/Description
Piezoelectric Sensor/Energy Harvester	Dual-function element; detects swallowing vibrations and converts mechanical stress to electrical charge [4] [13].
Energy Harvesting PMIC	Manages power extraction, conversion, and storage (e.g., models with integrated MPPT and low quiescent current) [61].
Hybrid Storage Unit	Combination of a Li-ion battery and a supercapacitor for balanced energy and power density [62].
Programmable Load Circuit	Emulates the dynamic power consumption profile of a sensor node (microcontroller, wireless transceiver).
Vibration Shaker Table	Simulates jaw and laryngeal movements with controlled frequency and amplitude.
Data Acquisition (DAQ) System	High-impedance input for measuring voltage, current, and power from the harvester and storage units.
Oscilloscope	For visualizing transient waveforms and signal integrity.

4.3 Procedure

• System Integration: Connect the piezoelectric harvester to the input of the PMIC. Connect the output of the PMIC to the HEES unit (battery and supercapacitor). Finally, connect the programmable load to the output of the HEES unit.
• Simulated Activity Profile: Program the vibration shaker table to produce a signal that mimics a standardized chewing and swallowing sequence, such as one based on the Test of Masticating and Swallowing Solids (TOMASS) [4]. The profile should include periods of activity (chewing bursts) and inactivity (pauses).
• Data Collection:
  • Simultaneously record the voltage and current at the output of the piezoelectric harvester, the output of the PMIC, and across the HEES unit using the DAQ system.
  • Calculate instantaneous and average power at each point.
  • Measure the voltage across the supercapacitor and battery to track state-of-charge over multiple activity cycles.
• Performance Metrics Calculation:
  • Total Harvested Energy: Integrate the power-over-time curve at the harvester output.
  • System Efficiency: (Useful energy delivered to the load / Total harvested energy) × 100%.
  • Load Operation Time: Duration for which the sensor node remains fully operational without any external energy input, starting from a fully charged HEES system.

The logical flow of the experimental setup and data acquisition process is outlined below.

Power Management Workflow for a Wearable Sensor

The following diagram illustrates the complete power management and task execution logic for a wearable chewing and swallowing detection sensor, from energy acquisition to application-level output.

Integrating advanced power management and energy harvesting strategies is not merely an enhancement but a fundamental requirement for the practical deployment of autonomous, wearable sensors for chewing and swallowing detection. The combination of highly efficient PMICs, capable of achieving over 80% efficiency from vibrational sources, with a Hybrid Electrical Energy Storage system creates a robust and sustainable power core. The experimental protocols outlined provide a framework for researchers to quantitatively validate and optimize these systems, ensuring they meet the demanding power profile of continuous physiological monitoring. This approach paves the way for long-term, unobtrusive health assessment tools that are critical for managing conditions like dysphagia.

The automatic detection of chewing and swallowing is a critical component in the development of objective monitoring systems for dietary intake and the screening of swallowing disorders (dysphagia). Research in this field, particularly within the context of a broader thesis on piezoelectric sensor applications, must account for two fundamental sources of data variability: bolus type (the consistency and volume of the substance being swallowed) and subject-dependent factors (individual anatomical and physiological differences). Effectively addressing this variability is paramount for creating robust, generalizable detection algorithms suitable for use in free-living conditions and diverse patient populations.

This application note provides a structured overview of the key factors influencing variability, summarizes quantitative findings from the literature, and offers detailed experimental protocols designed to systematically evaluate and mitigate these challenges in piezoelectric sensor-based research.

Quantifying the Impact of Variability

The performance of swallowing detection and characterization systems is significantly influenced by bolus properties and inter-subject differences. The following tables synthesize quantitative evidence from published studies.

Table 1: Impact of Bolus Type on Detection and Characterization Accuracy

Bolus Characteristic	Sensor Modality	Key Finding	Reported Performance	Citation
Food Type (Solid)	Audio-based (Throat Microphone)	Higher classification accuracy for dry foods (chips, nuts) compared to a sandwich or meat patty.	F-Measure: Chips=100%, Nuts=92.3%, Patty=85.2%	[28]
Food Texture (Solid)	sEMG (Sternocleidomastoid)	High accuracy in classifying solid bolus texture (hard, intermediate, soft).	Classification Accuracy: >99% (Random CV), >94% (Cross-Subject CV)	[63]
Bolus Volume (Liquid)	sEMG (Sternocleidomastoid)	Neural network-based regression can estimate liquid bolus volume.	R² = 0.88, RMSE = 0.2 ml (Min volume: 10 ml)	[63]
Liquid vs. Solid	sEMG & Wrist IMU Fusion	sEMG signal magnitude (MAM) differentiates fluid/swallow (low MAM) from bite/chew/swallow (high MAM).	Successful activity segmentation and classification	[63]

Table 2: Impact of Subject-Dependent Factors and Sensor Modality Performance

Factor / Modality	Description	Impact / Performance	Citation
Inter-Subject Variance	Significant individual differences in masticatory cycles during meals of fixed size.	Swallowing rates varied from 80 to 510 for a fixed meal.	[2]
Body Morphology	Sensors relying on laryngeal movement detection (e.g., strain sensors) are less reliable for obese subjects.	Submandibular fat pads inhibit reliable detection of swallows.	[10] [2]
Audio vs. Piezoelectric (Inertial)	Comparison of throat microphone vs. neck-mounted piezoelectric sensor for food classification.	Audio-based approach had significantly higher classification accuracy.	[28]
Power Consumption	Comparison of computational and power overhead between audio and piezoelectric-based systems.	Audio-based systems incur a power overhead approximately 30x greater than piezoelectric-based systems.	[28]

Experimental Protocols for Addressing Variability

To ensure your research on piezoelectric sensors for chewing and swallowing detection yields robust and generalizable results, the following experimental protocols are recommended.

Protocol 1: Systematic Bolus Variation

This protocol is designed to characterize the sensor's response across a wide range of bolus types.

• Participant Preparation: Recruit participants who meet the study's inclusion criteria (e.g., healthy adults, 18-65). Obtain informed consent. Position and secure the piezoelectric sensor on the participant's neck, typically on the cricothyroid region, using a medical-grade adhesive or a wearable collar.
• Bolus Selection:
  • Liquids: Use water and viscous liquids (e.g., nectar-thick, honey-thick) with volumes of 5 mL, 10 mL, and 20 mL.
  • Solids: Use foods representing a spectrum of textures:
    • Hard: Raw carrot sticks (3 g, 5 g)
    • Intermediate: Cheese cubes (3 g, 5 g)
    • Soft: Banana pieces (3 g, 5 g)
• Data Collection:
  • For each bolus type and volume, instruct the participant to place the bolus in their mouth and swallow naturally upon a verbal cue.
  • Record the piezoelectric sensor signal throughout the trial.
  • For solid foods, ensure the trial captures the entire sequence from ingestion to swallow.
  • Repeat each condition a minimum of 5 times to account for intra-meal variability.
• Reference Annotations: Simultaneously, use a manual scoring method (e.g., video recording or a second, validated modality like sEMG) to mark the precise timestamps of chewing bursts and swallows. This serves as the gold standard for training and validation [2].

The workflow for this protocol is outlined below.

Protocol 2: Accounting for Subject-Dependent Factors

This protocol focuses on evaluating and controlling for variability across a diverse participant cohort.

• Cohort Recruitment: Deliberately recruit a participant cohort that is heterogeneous in key aspects:
  • Body Mass Index (BMI): Include participants from normal weight, overweight, and obese categories.
  • Age: Include a range of ages, particularly if the target application involves the elderly.
  • Health Status: Include both healthy controls and individuals with conditions of interest (e.g., mild dysphagia).
• Baseline Data Collection: For each participant, record baseline signals:
  • Resting Swallows: Capture several saliva swallows.
  • Head Movement Artifacts: Record signals while the participant performs head turns, nods, and talking. This data is crucial for training classifiers to distinguish swallows from motion artifacts.
• Standardized Meal: Administer a standardized test meal (e.g., 200 mL of water, 5 saltine crackers) to all participants. Record the entire eating session using the piezoelectric sensor.
• Data Annotation and Analysis:
  • Manually annotate all chewing and swallowing events from the recorded sessions.
  • Perform analysis both on a per-subject basis (personalized models) and across the entire cohort (generalized models) to quantify performance degradation due to inter-subject variability.

The logical relationship for analyzing subject-dependent factors is as follows.

The Scientist's Toolkit: Research Reagent Solutions

Table 3: Essential Materials for Piezoelectric Sensor-Based Swallowing Research

Category	Item / Technique	Specific Function / Rationale	Citation
Core Sensor	Piezoelectric Sensor / Accelerometer	Detects mechanical vibrations and motions associated with laryngeal movement during swallowing. Placed on the cricothyroid region of the neck.	[28] [2] [64]
Reference Sensors	Throat Microphone	Captures acoustic signatures of swallows; provides a complementary signal modality for fusion or validation, though higher power consumption.	[13] [28]
	Surface Electromyography (sEMG)	Monitors muscular activity of the sternocleidomastoid or submental muscle group; provides a physiological ground truth for swallowing and chewing events.	[10] [63]
Signal Annotation	Manual Scoring Protocol (Video & Sensor)	Creates a gold-standard dataset by having trained raters annotate the onset/offset of chews and swallows using synchronized video and sensor signals. Critical for supervised learning.	[2]
Data Processing	Machine Learning / Deep Learning Models (SVM, LSTM, CNN, CTC)	Classifies sensor data into swallowing/chewing events and distinguishes between different bolus types. End-to-end models like hybrid CTC/attention can learn directly from weakly labeled data.	[65] [13] [63]
Experimental Materials	Standardized Bolus Samples	Liquids of varying viscosity and solids of varying texture (e.g., carrot, cheese, banana) to systematically evaluate the impact of bolus type on signal morphology.	[28] [63]

Benchmarking Performance and Clinical Translation

The validation of novel chewing and swallowing detection technologies against established clinical gold standards is a critical step in translating research into credible tools for scientific and clinical practice. For a thesis investigating piezoelectric sensors, constructing a robust validation framework is paramount. This document outlines application notes and experimental protocols for correlating data from a piezoelectric sensor system with Videofluoroscopic Swallowing Study (VFSS) and High-Resolution Manometry (HRM). These protocols are designed to provide researchers with a structured methodology to quantitatively assess sensor performance, establish convergent validity, and characterize the relationship between external sensor signals and internal physiological events.

Comparative Analysis of Gold-Standard Instrumental Assessments

A clear understanding of the clinical gold standards—VFSS and HRM—is essential for designing a meaningful validation study. The table below summarizes their primary characteristics, which directly inform the design of correlation experiments.

Table 1: Key Characteristics of Clinical Gold-Standard Assessments for Swallowing

Feature	Videofluoroscopic Swallowing Study (VFSS)	High-Resolution Manometry (HRM)
Primary Measured	Dynamic bolus flow, anatomical movement, and timing of swallowing events [66]	Spatiotemporal pressure data along the entire pharynx and esophagus [67]
Key Parameters	Bolus clearance, laryngeal penetration/aspiration, pharyngeal residue, hyoid excursion	Pressure, peristaltic wave velocity, sphincter function (e.g., Integrated Relaxation Pressure) [67]
Physiological Correlation for Piezoelectric Sensors	Timing of swallow initiation, gross laryngeal elevation, bolus passage sounds	Direct correlation with pressure waves generated by pharyngeal and esophageal peristalsis
Primary Diagnostic Use	Visualization of swallowing safety and efficiency [66]	Quantification of swallowing biomechanics and motility disorders [67]
Advantages	Direct visualization of bolus flow and aspiration; functional assessment	Highly quantitative; detailed pressure topography; identifies motility patterns
Limitations	Radiation exposure; qualitative analysis risk; limited soft tissue detail	Does not visualize aspiration; measures pressure but not bolus flow directly

Experimental Protocol for VFSS Correlation

This protocol is designed to validate piezoelectric sensor signals against the dynamic anatomical and bolus flow information provided by VFSS.

Research Reagent Solutions

Table 2: Essential Materials for VFSS Correlation Studies

Item	Function/Description
Piezoelectric Sensor System	The device under validation; should be calibrated for signal amplitude and frequency response.
Videofluoroscopic System	Provides real-time X-ray imaging; must be capable of recording for subsequent frame-by-frame analysis.
Barium Sulfate Preparations	Radiocontrast agent for visualizing boluses of different consistencies (e.g., thin liquid, nectar-thick liquid, pudding, solid) [66].
Synchronization Trigger	A device (e.g., light sensor, audio cue) to generate a simultaneous marker on both the VFSS video and sensor data stream.
Annotation Software	Software like ELAN for manual behavioral coding of swallowing events from video recordings [5].

Methodology

• Participant Preparation & Sensor Placement: After obtaining informed consent, securely attach the piezoelectric sensor(s) to the participant's neck skin, typically anterior to the sternocleidomastoid muscle at the level of the larynx, using a double-sided adhesive tape. Ensure the sensor has good skin contact for optimal signal acquisition.
• System Synchronization: Prior to data collection, initiate the synchronization trigger to place a clear, simultaneous time-locked marker on both the VFSS recording and the piezoelectric sensor data stream. This is critical for aligning the datasets during analysis.
• Data Acquisition: Instruct the participant to swallow a series of barium-coated boluses. The protocol should include multiple trials (e.g., 3-5) for each of the following consistencies: 5mL thin liquid, 5mL nectar-thick liquid, and 5mL pudding. Vary the order of consistencies to avoid bias.
• Video Annotation (Gold Standard): A trained analyst, blinded to the sensor data output, will review the VFSS recordings using annotation software. The following events should be marked with high temporal precision:
  • Swallow Initiation (SI): The frame where the bolus head passes the ramus of the mandible.
  • Laryngeal Vestibular Closure (LVC): The frame where the laryngeal vestibule is fully closed.
  • Upper Esophageal Sphincter Opening (UES-O): The frame of maximal UES opening.
  • Swallow Completion (SC): The frame where the bolus tail passes through the UES.
• Sensor Signal Analysis: Export the raw piezoelectric signal data. Apply signal processing techniques (e.g., band-pass filtering, rectification, smoothing) to identify and mark salient features in the signal that correspond to swallowing events.
• Correlation & Statistical Analysis:
  • Temporal Agreement: Calculate the time difference between the VFSS-annotated events (SI, LVC) and the corresponding features in the piezoelectric signal for each swallow. Report the mean difference, standard deviation, and limits of agreement (Bland-Altman analysis).
  • Detection Accuracy: For each swallow event, classify the sensor's detection as a true positive, false positive, or false negative against the VFSS benchmark. Calculate sensitivity, specificity, and accuracy.

The following workflow diagram illustrates the key steps in this protocol:

Diagram 1: VFSS Correlation Workflow

Experimental Protocol for HRM Correlation

This protocol is designed to correlate the piezoelectric sensor signal with the precise intraluminal pressure data obtained from HRM.

Research Reagent Solutions

Table 3: Essential Materials for HRM Correlation Studies

Item	Function/Description
High-Resolution Manometry System	A solid-state or water-perfused catheter system with ≥ 20 pressure sensors spanning from pharynx to stomach [67].
Piezoelectric Sensor System	The device under validation.
HRM Catheter	A catheter with closely spaced circumferential pressure sensors; impedance sensors can be integrated to simultaneously assess bolus flow.
Data Acquisition & Synchronization Software	Software provided by the HRM manufacturer, capable of recording and time-synchronizing external analog inputs (e.g., the piezoelectric signal).

Methodology

• Catheter & Sensor Placement: Following clinical guidelines, transnasally place the HRM catheter such that sensors span from the velopharynx to the proximal stomach. Simultaneously, attach the piezoelectric sensor to the neck skin, as described in the VFSS protocol.
• Hardware Synchronization: Connect the analog output of the piezoelectric sensor system to an auxiliary input channel on the HRM system. Use the HRM software to record both the internal pressure data and the external piezoelectric signal on a unified timeline.
• Swallowing Protocol: Conduct a series of swallow trials following the Chicago Classification protocol [67]. This should include at least ten 5mL single water swallows in a supine or upright position. To increase clinical relevance, adjunctive tests like multiple rapid swallows or a solid test meal (e.g., 2cm x 2cm bread) can be incorporated.
• HRM Data Analysis (Gold Standard): Using the HRM analysis software, identify key pressure events for each swallow:
  • Pharyngeal Pressure Peak: The maximum pressure in the pharyngeal region.
  • Upper Esophageal Sphincter (UES) Relaxation: The minimum UES pressure, characterized by the Integrated Relaxation Pressure (IRP) metric.
  • Peristaltic Wave Onset & Peak: The initiation and peak of the esophageal peristaltic wave.
• Signal Correlation Analysis:
  • Temporal Correlation: Measure the time difference between the onset of the pharyngeal pressure peak on HRM and the onset of the major high-frequency signal burst from the piezoelectric sensor.
  • Amplitude-Pressure Relationship: For each swallow, extract the peak amplitude or area under the curve of the processed piezoelectric signal. Perform correlation analysis (e.g., Pearson's correlation) between this sensor metric and the peak pharyngeal pressure measured by HRM.

The logical relationship between the measured signals and the analysis path is shown below:

Diagram 2: HRM Correlation Logic

Quantitative Data from Validation Studies

To establish the performance expectations for a novel sensor, the table below summarizes quantitative results from validation studies of existing technologies, which can serve as benchmarks.

Table 4: Performance Metrics from Selected Validation Studies

Technology / Study Focus	Validation Method	Key Performance Metrics	Relevance to Piezoelectric Validation
OCOsense Glasses (Facial Movement) [5]	Manual video coding of chewing	Chew Count: Strong correspondence with manual coding (r=0.955).Eating Detection: 81% sensitivity.Non-Eating Detection: 84% specificity.	Demonstrates high agreement possible for orofacial movements; a target for chewing detection validation.
AI-based Dysphagia Screening [1]	Clinical diagnosis (VFSS/FEES)	Accuracy: Ranged from 71.2% to 99%.Sensitivity: Ranged from 63.6% to 100%.AUROC: Ranged from 0.77 to 0.977.	Shows a wide performance range; highlights need for robust testing. Multimodal systems generally outperformed unimodal.
Flexible Balloon Sensor (Esophagus) [68]	Simulated peristalsis on ex vivo tissue	Impedance Detection: Could distinguish stressed segments and detect impedance changes.	Provides an example of validating a novel sensor against a controlled mechanical simulation.
Acoustic Swallowing Detection [13]	Manual annotation of audio signals	Model Performance: Used hybrid CTC/attention model with weak labels to detect swallowing and chewing side.	Illustrates an algorithmic approach for temporal event detection from a continuous signal, applicable to piezoelectric data.

In the development and validation of piezoelectric sensor systems for chewing and swallowing detection, quantitative performance metrics are indispensable. These metrics provide researchers with an objective basis for evaluating how well a sensor or algorithm can distinguish between true physiological events and non-events. The core metrics—Accuracy, Sensitivity, Specificity, and the Area Under the Receiver Operating Characteristic Curve (AUC)—each offer a distinct perspective on model performance. For researchers in drug development and clinical sciences, understanding these metrics is crucial for assessing the potential of wearable technology to provide reliable, long-term swallowing assessment outside clinical settings, thereby facilitating earlier detection and management of dysphagia [69] [70].

The following table summarizes the reported performance ranges of sensor-based technologies for swallowing detection, illustrating the current state of the field:

Table 1: Reported Performance Ranges of Sensor-Based Swallowing Detection Technologies

Performance Metric	Reported Range in Literature	Contextual Notes
Accuracy	71.2% to 99% [69]>90% to <60% [70]	Wide variability; some systems report high performance while others show poor accuracy [70].
Sensitivity	63.6% to 100% [69]	Measures the ability to correctly identify true swallowing events.
Specificity	Not Quantified	While critical, specificity values were not consistently reported across the reviewed literature.
AUC	0.77 to 0.977 [69]	Provides an aggregate measure of separability between swallow and non-swallow classes.

Detailed Experimental Protocols for Swallowing Detection

Sensor Fabrication and Data Collection

A critical step in this research domain involves the fabrication of conformable sensors and the standardized collection of swallowing signal data.

Protocol 1: Fabrication of a Flexible Piezoelectric Sensor Patch This protocol is adapted from studies utilizing Aluminum Nitride (AlN) as the piezoelectric material [11].

• Substrate Preparation: Begin with a soft Kapton polyimide foil substrate (25 μm thick) [11].
• Thin-Film Deposition: Using standard microfabrication techniques like sputtering, deposit the following layers onto the substrate:
  • A 120 nm AlN interlayer (AlN-IL).
  • A 200 nm Molybdenum (Mo) bottom electrode.
  • A 1 μm thick piezoelectric AlN layer.
  • A 200 nm Molybdenum (Mo) top electrode [11].
• Patterning & Packaging: Pattern the electrode layers via photolithography. Employ a precision 3D-printing system to create a sealed package and implement metal contacts on the electric pads, ensuring the device remains lightweight (<2 g) and flexible [11].
• System Integration: Integrate the fabricated sensor patch with a custom charge amplifier and a low-power, wireless data transmission module (e.g., Bluetooth) for connection to a smartphone or other data acquisition system [11].

Protocol 2: Data Collection for Swallowing and Non-Swallowing Events This protocol ensures the collection of a robust dataset for model training and testing [69] [71] [70].

• Participant Recruitment: Recruit a cohort of healthy adult volunteers. Studies have included up to 20 participants for initial validation, though larger sample sizes are needed to reduce the risk of bias [71] [70].
• Sensor Placement:
  • For laryngeal movement/vibration: Attach the sensor to the neck skin in the region of the laryngeal prominence (Adam's apple) using medical tape, ensuring good conformability to minimize motion artifact [11] [70].
  • For jaw motion (chewing): Place a piezoelectric strain sensor immediately below the outer ear, where jaw motion causes changes in skin curvature [71].
• Task Protocol: Record signals while participants perform a series of tasks to capture both target events and confounding signals:
  • Swallowing Tasks: Administer specific volumes of water or food (e.g., 3 mL in the Modified Water Swallowing Test) [11] [70].
  • Dry Swallowing: Ask the participant to swallow their saliva spontaneously [70].
  • Non-Swallowing Tasks: Include activities such as quiet sitting, reading, talking, and yawning to collect negative samples [71] [70].
• Data Acquisition: Sample the analog sensor signal (e.g., at 100 Hz), buffer it with a high-input-impedance operational amplifier, and digitize it using a data acquisition module [71].

The workflow for this experimental setup and data processing can be visualized as follows:

Signal Processing and Model Validation

Once data is collected, signal processing and machine learning are used to build a detection model.

Protocol 3: Signal Processing and Feature Extraction

• Segmentation: Divide the continuous signal into fixed-length, non-overlapping epochs (e.g., 30-second windows) [71].
• Pre-processing: Apply filters to remove high-frequency RF noise and eliminate any high-voltage spikes [71].
• Feature Extraction: For each epoch, extract a comprehensive set of time-domain and frequency-domain features. One methodology extracted 250 initial features per epoch [71].
• Feature Selection: Implement a feature selection procedure (e.g., forward feature selection) to identify the most critical features for classification, reducing dimensionality and mitigating overfitting. Studies have identified between 4 and 11 features as being most critical [71].

Protocol 4: Model Training and Performance Validation

• Model Selection: Employ a Support Vector Machine (SVM) classifier, which is the most common model in this field. Alternatively, explore deep learning approaches, which are emerging in recent years [69] [71].
• Dataset Partitioning: Split the epoch-based dataset into training and testing sets. To maximize data usage, perform a 20-fold cross-validation [71].
• Performance Calculation: Calculate key metrics on the testing set for each fold:
  • Accuracy: (True Positives + True Negatives) / Total Epochs.
  • Sensitivity (Recall): True Positives / (True Positives + False Negatives).
  • Specificity: True Negatives / (True Negatives + False Positives).
  • AUC: Calculate the Area Under the ROC Curve, which plots the True Positive Rate (Sensitivity) against the False Positive Rate (1 - Specificity) at various threshold settings [69] [71].
• Reporting: Report the mean and standard deviation of each performance metric across all cross-validation folds to ensure a robust evaluation [71].

The pathway from raw signal to performance metrics is outlined below:

The Scientist's Toolkit: Research Reagent Solutions

Successful research in this area relies on a suite of essential materials and tools. The following table details key components of the experimental toolkit.

Table 2: Essential Research Materials and Tools for Piezoelectric Swallowing Detection

Item Name	Function/Description	Specific Examples & Notes
Piezoelectric Material	Converts mechanical strain (from larynx movement) into an electrical signal.	Aluminum Nitride (AlN): Biocompatible, can be deposited on soft substrates [11].Polyvinylidene Fluoride (PVDF): A flexible polymer used in commercial sensors [71].
Flexible Substrate	Provides a soft, conformable base for the sensor, enabling comfort and good skin contact.	Kapton Foil: A polyimide film known for its structural and mechanical properties at thin dimensions (e.g., 25 μm) [11].
Data Acquisition System	Converts the analog sensor signal into a digital format for computer analysis.	Components include a buffering op-amp (e.g., TLV-2452) and a digitizer (e.g., USB-1608FS) sampling at 100 Hz [71].
Machine Learning Framework	Provides the environment for developing and testing classification algorithms.	Support Vector Machine (SVM): The most commonly used model in current literature [69] [71].Deep Learning: Emerging approach for more complex pattern recognition [69].
Validation Software	Tools to calculate and visualize key performance metrics from model predictions.	Software capable of generating confusion matrices and plotting Receiver Operating Characteristic (ROC) curves to calculate AUC [69].

Critical Analysis of Methodological Rigor and Bias

When interpreting the performance metrics in Table 1, it is vital to consider the methodological context. A recent systematic review highlighted significant risks of bias in this field, particularly concerning patient selection and the index test [69]. A primary concern is that none of the reviewed studies conducted external validation or domain adaptation testing, meaning their reported high performance may not generalize well to real-world, unseen data [69]. Furthermore, many studies suffer from small sample sizes and imbalanced class sizes, which can artificially inflate accuracy estimates [70]. Researchers must therefore critically evaluate not only the reported metrics but also the rigor of the validation methodology. Future work should prioritize independent external validation cohorts and standardized assessment protocols to ensure that performance metrics translate into clinical utility [69] [70].

The automatic detection of chewing and swallowing is a critical research area for monitoring dietary intake, managing chronic health conditions, and diagnosing swallowing disorders like dysphagia. Wearable sensor technologies have emerged as promising, non-invasive alternatives to clinical gold standards like videofluoroscopy. Among these, systems based on piezoelectric sensors, acoustic microphones, and accelerometers are some of the most widely investigated. This application note provides a comparative analysis of these technologies, presenting structured performance data, detailed experimental protocols, and implementation frameworks to assist researchers in selecting and deploying the appropriate sensing modality for their specific applications.

The operating principles of these systems are fundamentally different. Piezoelectric sensors typically measure mechanical vibrations on the skin surface resulting from swallowing and chewing activity, often configured as part of a "smart necklace" [28]. Acoustic-based systems use microphones (throat or external) to capture the sound waves produced by these activities [28] [13]. Accelerometer-based systems detect vibrations and movements of the laryngeal structures, a method also known as high-resolution cervical auscultation [72] [70] [42]. Some research explores multimodal sensing that combines these technologies to improve overall accuracy [1] [73].

Table 1: Comparative Performance of Swallowing Detection Technologies

Technology	Reported Accuracy Range	Key Strengths	Key Limitations	Power Consumption
Piezoelectric Sensing	75.3% - 79.4% (F-Measure for food classification) [28]	Low power requirements; Directly measures skin motion [28]	Lower classification accuracy compared to audio; Requires firm skin contact [28]	Low (Approx. 1.8 mW at 100 Hz) [28]
Acoustic Sensing	88.5% - 91.3% (F-Measure for food classification) [28]	High classification accuracy for different food types [28]	Higher power consumption; Susceptible to ambient noise [28] [73]	High (Approx. 27.2 mW at 8 kHz) [28]
Accelerometer (Single)	Up to >90% (in some studies) [72] [70]	Low cost; Robust to acoustic noise [72] [42]	Performance varies widely (<60% to >90%); Sensitive to motion artifacts [72] [70]	Moderate
Multimodal / Advanced Systems	Up to 95.96% (STVS with ensemble deep learning) [73]	High robustness and accuracy; Resistant to ambient noise [73]	Increased system complexity and form factor [1] [73]	Varies (typically Moderate-High)

Detailed Experimental Protocols

To ensure reproducible and valid results, standardized experimental protocols for data collection are essential. The following sections outline established methodologies for evaluating these sensing technologies.

Protocol for Comparative Food Classification

This protocol is adapted from a study directly comparing piezoelectric and acoustic methods [28].

• Objective: To evaluate the performance of piezoelectric and acoustic sensors in classifying different types of swallowed foods.
• Subjects: 20 subjects with an age range of 22-40 years [28].
• Sensor Configuration:
  • Piezoelectric Sensor: A custom sensor placed on the lower part of the neck (laryngeal prominence) to monitor skin vibrations from swallows. The sensor should have firm skin contact.
  • Acoustic Sensor: A commercial throat microphone (e.g., Hypario Throat Microphone) placed loosely around the neck in the lower collarbone area.
• Data Collection Procedure:
  • Experiment I: Subjects sequentially consume water, a sandwich, and chips.
  • Experiment II: Subjects sequentially consume nuts, chocolate, and a meat patty.
  • For each food type, data from both sensors are collected simultaneously.
  • Swallows are manually annotated by researchers through visual observation and palpation to create a ground truth dataset.
• Data Analysis:
  • Feature Extraction: For audio signals, use a tool like openSMILE to extract a large set of acoustic features, including MFCCs, PLP, and spectral features [28].
  • Classification: Use a machine learning classifier (e.g., Random Forests) with leave-one-out cross-validation to classify the food type from the sensor data and report precision, recall, and F-Measure.

Protocol for Swallow Detection in Solid Food Consumption

This protocol leverages a neck-worn electronic stethoscope (NWES) to automate the Test of Masticating and Swallowing Solids (TOMASS) [4].

• Objective: To objectively measure masticatory and swallowing parameters during solid food consumption.
• Subjects: Healthy adults without dysphagia or cognitive impairments. A large sample size (e.g., n=123) is recommended to account for age and gender variations [4].
• Sensor Configuration:
  • Device: A neck-worn electronic stethoscope (NWES) with a piezoelectric vibration sensor is positioned on the anterior neck between the C2 and C5 vertebrae [4].
  • Auxiliary Device: A smartphone camera to record a close-up video of the consumption process for ground truth validation.
• Data Collection Procedure:
  • Participants don the NWES.
  • They are given two standard crackers (e.g., 3g Nabisco Premium Crackers).
  • Participants are instructed to eat one cracker at a time at their normal pace and verbally indicate "Finished" upon completion.
  • The NWES records swallowing sounds while the smartphone video records the entire process.
• Data Analysis & Measured Parameters:
  • Synchronization: Manually synchronize the audio and video recordings using software (e.g., ELAN) by aligning the "Finished" utterance.
  • Parameter Extraction:
    • Discrete Bite Count: Counted from the video recording.
    • Swallow Count: Counted by combining video observation and the audio waveform from the NWES.
    • Oral Processing and Swallowing Time (OPST): The duration from the first bite sound to the verbal indication of completion.
    • First OPST (1st-OPST): The duration from the first bite sound to the onset of the first swallow sound.

Protocol for Throat-Related Event Classification

This protocol uses a specialized wearable sensor and an advanced deep learning model for classifying multiple throat activities [73].

• Objective: To automatically detect and classify various throat-related events (swallowing, coughing, speaking, throat clearing) in a continuous data stream.
• Subjects: A diverse group of subjects (e.g., n=32) across ages and genders [73].
• Sensor Configuration:
  • Device: A Soft Skin-Attachable Throat Vibration Sensor (STVS). The flexible sensing part is placed on the neck skin above the laryngeal prominence, while the controller is positioned on the side of the neck [73].
• Data Collection Procedure:
  • Subjects wear the STVS.
  • They perform five repetitions of each event: coughing, speaking (in multiple languages for generalization), swallowing, and throat clearing.
  • A rest interval of 2 seconds is maintained between each event.
  • Data is stored on the sensor and transmitted via Bluetooth Low Energy (BLE) to a database.
• Data Analysis:
  • Data Segments: The continuous data stream is segmented into shorter windows (e.g., 625 ms).
  • Model Training: An ensemble-based deep learning model is trained on multi-modal acoustic features (both time and frequency domains) extracted from the data.
  • Performance Evaluation: The model is evaluated using a separate test set, reporting classification accuracy and area under the ROC curve.

The Scientist's Toolkit: Essential Research Reagents & Materials

Table 2: Key Materials and Equipment for Swallowing Detection Research

Item	Specification / Example	Primary Function in Research
Piezoelectric Sensor	Custom or commercial vibration sensor (e.g., for smart necklace) [28]	Detects mechanical skin vibrations on the neck associated with swallowing and laryngeal movement.
Throat Microphone	Commercial microphone (e.g., Hypario Throat Microphone) [28]	Captures acoustic signals of swallows and chews directly from the throat, minimizing ambient noise.
Neck-Worn E-Stethoscope	Custom device with piezoelectric sensor [4]	A specialized form factor for consistent placement on the neck to capture swallowing sounds.
Accelerometer (IMU)	Single or dual inertial measurement units [42] [63]	Measures neck surface vibrations and movements; dual IMUs can help cancel motion artifacts.
sEMG Sensor	Wireless single-electrode system (e.g., BITalino) [63]	Records muscular activity from the sternocleidomastoid or submental muscle group during swallowing.
Synchronization Camera	Smartphone camera or similar [4]	Provides video ground truth for manual annotation of bites, chews, and swallows.
Annotation Software	ELAN [4]	Software for manually annotating and synchronizing video and audio data to create labeled datasets.
Feature Extraction Tool	openSMILE toolkit [28]	Extracts a comprehensive set of acoustic features from audio signals for machine learning.

Implementation Workflow and System Trade-offs

The general workflow for deploying a swallowing detection system involves several key stages, from signal acquisition to final classification. The following diagram illustrates this process and the critical choices at each stage.

Diagram 1: Generalized workflow for implementing a swallowing detection system, highlighting key decision points at each stage.

The choice of sensor technology involves inherent trade-offs, primarily between detection accuracy and system power consumption, as quantified in the following visualization.

Diagram 2: The fundamental trade-off between classification accuracy and system power consumption for different sensing modalities.

The selection of an optimal sensing technology for chewing and swallowing detection is dictated by the specific research requirements. Piezoelectric systems offer a low-power solution suitable for long-term monitoring where power efficiency is paramount, though at the cost of lower classification accuracy. Acoustic-based systems provide superior accuracy for distinguishing between different foods and swallow types but demand higher computational resources and are susceptible to environmental noise. Accelerometer-based systems strike a balance, offering robustness to acoustic noise, though performance can be variable and sensitive to motion artifacts. Emerging research points toward multimodal sensing and advanced soft, skin-attachable sensors combined with sophisticated deep learning models as the future direction for achieving high accuracy and robustness in real-world environments. Researchers are encouraged to use the provided protocols and tables as a foundation for designing rigorous and reproducible experiments in this rapidly evolving field.

Assessing Multimodal Approaches for Improved Diagnostic Accuracy

The quantitative assessment of masticatory and swallowing function is critical for diagnosing and managing conditions like dysphagia, which is associated with severe complications including aspiration pneumonia, malnutrition, and diminished quality of life [4] [11]. Traditional diagnostic methods, such as the Test of Masticating and Swallowing Solids (TOMASS), provide valuable insights but are often limited by operator dependency, subjective interpretation, and lack of objectivity [4]. Similarly, gold-standard techniques like Flexible Endoscopic Evaluation of Swallowing (FEES) and videofluoroscopic swallow study (VFSS), while accurate, are invasive, require significant expertise, and are unsuitable for frequent monitoring [11] [20].

Multimodal approaches that integrate various sensing technologies with artificial intelligence (AI) are emerging as powerful solutions to overcome these limitations. By combining complementary data sources—such as cervical auscultation sounds, piezoelectric sensor signals, and video analysis—these systems enable automated, objective, and precise quantification of swallowing and chewing events. This application note details the experimental protocols and key performance data for implementing such multimodal systems to enhance diagnostic accuracy in swallowing and masticatory function assessment.

Key Research Reagent Solutions

The following table catalogs essential materials and technologies used in advanced swallowing and chewing detection research.

Table 1: Essential Research Reagents and Technologies

Item Name	Function/Description	Key Features & Applications
Neck-worn Electronic Stethoscope (NWES) [4]	A contact microphone sensor positioned on the neck to collect swallowing sound data.	Enables automated detection and counting of swallows via deep learning-based sound analysis; used for objective TOMASS parameter measurement.
Flexible AlN Piezoelectric Sensor [11]	An ultrathin, compliant patch that converts laryngeal movement into a defined electrical signal.	Lightweight (<2 g), biocompatible; detects skin deformation from hyoid bone movement during swallowing for non-invasive monitoring.
Throat Microphone [28] [13]	An acoustic sensor placed on the neck to record audio signals associated with chewing and swallowing.	Captures high-fidelity eating sounds; used with ML classifiers for detecting and differentiating food intake types.
Hybrid CTC/Attention Model [13]	An end-to-end deep learning model for sequence recognition tasks.	Automatically detects and classifies eating behaviors (e.g., left/right chewing, swallowing) from sound data using weak labels.
FEES-CAD System [20]	An AI-assisted computer-aided diagnosis system for analyzing FEES video recordings.	Uses CNN to segment and classify aspiration, penetration, and residue in FEES videos at expert-level performance.

Performance Comparison of Sensing Modalities

Research directly compares the efficacy of different sensing modalities for monitoring dietary intake and swallowing.

Table 2: Comparative Performance of Audio vs. Piezoelectric-Based Swallow Sensing

Parameter	Audio-Based Detection (Throat Microphone) [28]	Piezoelectric-Based Inertial Sensing [28]
Primary Measured Signal	Acoustic waves from chewing and swallowing	Mechanical skin movement/vibration in the neck
Key Strengths	High classification accuracy for different food types	Lower power consumption; simpler data processing
Reported F-Measure (Exp I)	91.3%	75.3%
Reported F-Measure (Exp II)	88.5%	79.4%
Computational/Power Overhead	Higher (due to higher sample rates)	Lower
Ideal Use Case	Accurate identification and classification of food types	Long-term, energy-efficient monitoring of swallow count

Detailed Experimental Protocols

Protocol 1: Semi-Automated TOMASS Using a Neck-Worn Electronic Stethoscope

This protocol outlines the procedure for objectively measuring TOMASS parameters using a neck-worn electronic stethoscope (NWES), as validated in a study with 123 healthy adults [4].

1. Equipment Setup: - Neck-worn Electronic Stethoscope (NWES): Position the device around the anterior neck, ensuring the contact sensor sits between the C2 and C5 vertebrae. - Smartphone: Connect the NWES to a smartphone (e.g., via wired connection) to record audio data. Use the smartphone's camera to capture a close-up video of the cracker consumption process.

2. Test Execution: - Participant Preparation: Instruct the participant to sit comfortably. Provide two standard crackers (e.g., 3g each, Nabisco Premium Crackers). - Data Recording: For each cracker, instruct the participant to eat normally and to verbally indicate "Finished" upon completion. Simultaneously record swallowing sounds via the NWES and the visual feeding process via the smartphone video camera.

3. Data Reduction and Analysis: - Synchronization: Manually synchronize the audio and video recordings in analysis software (e.g., ELAN) by aligning the audio waveform with the video frame of the "Finished" utterance. - Parameter Extraction: - Discrete Bite Count: Count the total number of bites for a single cracker from the video recording. - Swallow Count: Count the number of swallows required to finish a cracker using both video and the synchronized audio waveform. - Oral Processing and Swallowing Time (OPST): Calculate the duration from the initial sound of biting the cracker to the onset of the "Finished" utterance. - First OPST (1st-OPST): Calculate the duration from the initial bite sound to the onset of the first swallow's sound.

Protocol 2: Detection of Swallowing Sounds vs. Laryngeal Movement

This protocol describes a methodology for comparing the performance of audio-based detection and flexible piezoelectric sensing [11] [28].

1. Sensor Configuration: - Audio-Based Setup: Place a commercial throat microphone loosely on the lower part of the neck (near the collarbone) to capture acoustic signals. - Piezoelectric-Based Setup: Affix a flexible AlN piezoelectric sensor to the skin of the neck, in the region of the laryngeal prominence, to detect mechanical movements.

2. Data Collection: - Participant Tasks: Recruit participants to consume a variety of test foods and liquids (e.g., water, sandwich, chips, nuts). Data should be collected using both sensors simultaneously. - Signal Acquisition: For the audio signal, use a high sample rate (e.g., sufficient for audio processing). For the piezoelectric sensor, a lower sample rate can be used as it captures lower-frequency mechanical motions.

3. Signal Processing and Classification: - Feature Extraction (Audio): Use a toolkit like openSMILE to extract a large set of acoustic features (e.g., MFCC, PLP, spectral features, voice quality) from the audio signals. - Classification: Train machine learning classifiers (e.g., Random Forests) on the extracted features to distinguish between different food types or to detect swallowing events. - Piezoelectric Signal Analysis: Analyze the signal from the piezoelectric sensor, which appears as a well-defined electrical waveform corresponding to the laryngeal excursion. The signal can be used to count swallows and characterize swallow duration.

Protocol 3: AI-Assisted Detection of Aspiration and Penetration in FEES Videos

This protocol leverages a deep learning system (FEES-CAD) to achieve expert-level automated analysis of FEES recordings [20].

1. Data Preparation: - Video Acquisition: Collect de-identified FEES video recordings from patients according to standard clinical procedures. - Expert Annotation: Have laryngologists with substantial experience manually annotate video frames. Labels should include anatomical structures (vocal fold, subglottis) and pathological events (aspiration, penetration, residue in vallecula/hypopharynx).

2. Model Training: - Network Architecture: Implement a Convolutional Neural Network (CNN) with an attention mechanism. The network should be designed for semantic segmentation and classification tasks. - Training Process: Train the network on a large set of annotated images (e.g., 25,000+ frames from hundreds of patient videos) to segment the objects of interest and classify the presence of aspiration, penetration, and residue.

3. Model Testing and Validation: - Performance Evaluation: Test the trained model on a separate set of FEES videos. Compare the AI's classifications (e.g., for aspiration, penetration) against the assessments of expert laryngologists. - Metrics: Calculate performance metrics including accuracy, sensitivity, specificity, and Dice similarity coefficient for segmentation tasks.

Experimental Workflow Visualization

The following diagram illustrates the integrated workflow for a multimodal swallowing assessment system, synthesizing elements from the cited protocols.

Analyzing Real-World Applicability and Domain Adaptation

The development of automated systems for chewing and swallowing detection represents a significant advancement in healthcare monitoring, particularly for managing conditions like dysphagia. While laboratory results for piezoelectric sensor-based systems and other wearable technologies show remarkable accuracy, their transition to reliable clinical use hinges on addressing two fundamental challenges: real-world applicability and domain adaptation. This analysis examines the current technological landscape, quantifies the performance gap between controlled and real-world environments, and provides detailed protocols for validating these systems for use in diverse, uncontrolled settings. Framed within broader thesis research on piezoelectric sensors, this document serves as a practical guide for researchers and drug development professionals aiming to build robust, clinically viable monitoring tools.

Current State of Sensor-Based Deglutition Detection

The field of computational deglutition has seen rapid growth, leveraging various sensor modalities and artificial intelligence (AI) models to screen for swallowing disorders [1] [69].

Predominant Sensing Modalities and AI Models

Systematic reviews reveal a focus on acoustic and vibratory signals. A 2025 scoping review of 24 studies found that acoustic signals (54%) and vibratory signals (38%) are the primary data sources for AI-based dysphagia screening [1] [69]. Piezoelectric sensors, which convert mechanical strain from laryngeal movement into quantifiable electrical signals, are a prominent vibratory technology [11]. While 75% of studies use a single modality, a emerging trend (25% of studies) employs multimodal approaches (e.g., combining acoustic and vibratory sensors), which generally demonstrate superior performance compared to unimodal systems [1] [69].

Machine learning, particularly Support Vector Machines (SVM), dominates current models. SVM is the most common AI model, featured in 62% of the reviewed studies [1] [69]. Deep learning approaches, while less common (12% of studies), are emerging in recent research. These models are typically used for per-individual classification (75% of studies) rather than analyzing individual swallow events [1] [69].

The following table summarizes the reported performance metrics of current AI-based instrument screening tools, highlighting their potential and the variability in their development.

Table 1: Performance Metrics of AI-Based Dysphagia Screening Tools (Based on a Review of 24 Studies)

Performance Metric	Reported Range	Notes
Accuracy	71.2% to 99%	Wide variability indicates dependence on specific experimental conditions [1] [69].
Area Under the Curve (AUC)	0.77 to 0.977	Indicates good to high classification ability in tested environments [1] [69].
Sensitivity	63.6% to 100%	Some systems show high success in identifying dysphagia cases [1] [69].
Diagnostic Odds Ratio (DOR)	Pooled DOR of 21.5	For wearable technology in identifying aspiration, indicating strong potential for clinical detection [69].

The Challenge of Real-World Applicability and Domain Adaptation

Despite promising performance in controlled studies, a critical analysis reveals significant methodological limitations that hinder the deployment of these technologies in clinical and home settings.

Identified Gaps in Methodological Rigor

A quality assessment of studies shows a high risk of bias, particularly in patient selection (unclear in 75% of studies) and index test (unclear in 96% of studies) [1] [69]. Most critically, the modeling domain was found to have a high risk of bias in 54% of the studies [1] [69]. A key finding from the review is that no studies conducted external validation or domain adaptation testing [1] [69]. This means the models are validated on data from the same source population (e.g., same clinic, same protocol), failing to demonstrate performance on data from new populations, different sensor batches, or varied environmental conditions.

Specific Domain Shift Challenges

For piezoelectric sensor systems, domain shifts can occur from multiple sources, which must be considered during experimental design:

• Physiological Variability: Anatomical differences (neck size, skin thickness), speech and swallowing patterns, and comorbid conditions can alter signal characteristics [11].
• Sensor and Hardware Variability: Differences between sensor batches, slight variations in placement on the neck, and adhesion quality can introduce signal drift or noise [11].
• Behavioral and Environmental Context: Dietary variations (bolus type and viscosity), physical activity, ambient noise (for acoustic systems), and user compliance in daily life create a signal environment far more complex than the laboratory [74].

Experimental Protocols for Assessing Applicability & Adaptation

To bridge the gap between laboratory performance and real-world utility, the following detailed protocols are proposed.

Protocol 1: Multi-Center External Validation

This protocol evaluates a pre-trained model's robustness across diverse, independent populations.

1. Objective: To assess the generalizability of a chewing/swallowing detection model by testing it on data collected from new clinical sites and participant demographics not included in the training set.

2. Research Reagent Solutions: Table 2: Key Materials for Multi-Center Validation

Item	Function/Explanation
Pre-trained AI Model	A model (e.g., SVM, Deep Learning) trained on a primary dataset. Its parameters are frozen for this validation.
Standardized Sensor Kit	Pre-configured packages containing the piezoelectric sensor, amplifier, and data logger to minimize hardware variability across sites [11].
Validated Reference Standard	The gold-standard method for swallow event labeling, such as Videofluoroscopic Swallow Study (VFSS) or Fiberoptic Endoscopic Evaluation of Swallowing (FEES) [1] [69].
Data Harmonization Scripts	Code (e.g., in Python/R) to normalize sampling rates, filter ranges, and signal amplitudes from different data acquisition systems.

3. Methodology:

• Site Selection: Recruit at least three independent clinical sites with varying patient populations (e.g., general geriatric, post-stroke, neurodegenerative).
• Data Collection: At each site, collect synchronized data from the piezoelectric sensor system and the reference standard (VFSS/FEES) during standardized swallowing tasks.
• Blinded Analysis: Apply the pre-trained model to the sensor data from each site. The model's predictions are compared to the reference standard labels without any model retraining.
• Performance Comparison: Calculate performance metrics (Accuracy, Sensitivity, Specificity, AUC) separately for each site's data and compare them to the model's original validation performance.

4. Analysis: A significant drop (e.g., >10%) in performance at external sites indicates poor generalizability and highlights the need for domain adaptation strategies.

Protocol 2: Unsupervised Domain Adaptation (UDA) for Sensor Data

This protocol outlines a method to adapt a model to a new target domain using only unlabeled data from that domain.

1. Objective: To improve model performance on data from a new target domain (e.g., a new hospital, a different sensor version) by leveraging unlabeled data from that domain, thus avoiding costly re-labeling.

2. Research Reagent Solutions: Table 3: Key Materials for Domain Adaptation

Item	Function/Explanation
Labeled Source Data	The original, fully-annotated dataset (e.g., Dataset A) used to train the base model.
Unlabeled Target Data	New data from the target domain (Dataset B) without swallow event labels.
Domain Adaptation Algorithm	Software implementation of algorithms such as Domain-Adversarial Neural Networks (DANN) or Deep CORAL, designed to learn domain-invariant features.
Feature Extraction Pipeline	Computational methods to transform raw piezoelectric signals into discriminative features (e.g., wavelet coefficients, statistical moments).

The following workflow diagram illustrates the core logic of a Domain-Adversarial Neural Network (DANN), a common UDA approach.

Diagram 1: Domain-Adversarial Neural Network (DANN) Workflow

3. Methodology:

• Base Model Training: Train an initial model on the labeled source data.
• Adapter Training: The model is further trained using a combination of labeled source data and unlabeled target data. The adaptation algorithm (e.g., DANN) works to simultaneously achieve two goals:
  • Minimize Label Prediction Loss: Ensure the model remains accurate at classifying swallows/chews on the source data.
  • Maximize Domain Confusion: Make the feature extractor produce features that are indistinguishable between the source and target domains, forcing it to learn domain-invariant representations.
• Validation: Test the adapted model on a separate, labeled test set from the target domain to measure performance improvement.

The Scientist's Toolkit: Research Reagent Solutions

A selection of essential materials and computational tools is critical for conducting rigorous research in this field.

Table 4: Essential Research Reagents and Tools

Category / Item	Specific Example / Technology	Function in Research
Sensor Technology
Flexible Piezoelectric Sensor	Aluminum Nitride (AlN) on Kapton substrate [11]	Conforms to neck skin, converts laryngeal movement into electrical signal with high strain resolution and low anatomical obstruction.
Signal Acquisition & Hardware
Charge Amplifier & Bluetooth Module	Custom-designed circuit [11]	Conditions the weak piezoelectric signal and enables wireless data transmission to a smartphone or computer for processing.
Reference Standard Equipment	Videofluoroscopic Swallowing Study (VFSS) [1] [69]	Provides gold-standard, frame-by-frame visualization of the swallowing mechanism for accurate labeling of sensor data.
Computational & Modeling Tools
Machine Learning Library	Scikit-learn (Python)	Provides implementations of classic models like Support Vector Machines (SVM) for initial classification tasks [1] [69].
Deep Learning Framework	PyTorch or TensorFlow	Enables building and training complex models like Transformers or Hybrid CTC/Attention networks for temporal event detection [13].
Domain Adaptation Library	DANN or CORAL implementations	Provides pre-built modules and functions for implementing domain adaptation algorithms, reducing development time.

The path from a high-accuracy laboratory model to a clinically useful tool for chewing and swallowing detection is fraught with challenges related to domain shift and real-world variability. Current systems, while promising, suffer from a lack of external validation. By adopting rigorous experimental protocols like multi-center validation and unsupervised domain adaptation, researchers can systematically quantify and address these limitations. Integrating these practices into the development lifecycle is paramount for creating piezoelectric sensor-based systems that are not only intelligent but also robust and reliable enough for real-world clinical and remote monitoring applications, ultimately improving patient care and advancing drug development research.

Conclusion

Piezoelectric sensor technology presents a transformative approach for objective chewing and swallowing detection, addressing critical limitations of current clinical standards. The integration of advanced materials, sophisticated data analytics, and wearable design is paving the way for continuous, long-term monitoring outside specialized clinics. Future progress hinges on overcoming methodological biases identified in current studies, particularly through robust external validation. The trajectory points towards intelligent, multimodal systems capable of dissecting the multifaceted nature of swallowing disorders, ultimately enabling personalized rehabilitation strategies and improved drug delivery assessment tools for researchers and clinicians.

References

• 1. Current Technological Advances in Dysphagia Screening [https://www.jmir.org/2025/1/e65551]
• 2. Non-invasive monitoring of chewing and swallowing for ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC2582220/]
• 3. Experimental Characterization of Optimized Piezoelectric ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC8587679/]
• 4. Automated Assessment of Test of Masticating and ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12624163/]
• 5. Validation of OCOsense glasses to detect chewing behaviour [https://www.sciencedirect.com/science/article/abs/pii/S0950329325003611]
• 6. Lead-free piezoelectric material converts motion to power ... [https://www.rdworldonline.com/lead-free-piezoelectric-material-converts-motion-to-power-without-lead/]
• 7. Piezoelectric Polymer Nanocomposite: Fabricating ... [https://www.sciencedirect.com/science/article/pii/S2666978125000959]
• 8. Experimental investigation of sensitivity changes during ... [https://www.nature.com/articles/s41598-025-16973-8]
• 9. Heartbeat electro-language: Exploring piezoelectric ... [https://www.sciencedirect.com/science/article/pii/S2667141725000667]
• 10. Wearable, epidermal devices for assessment of swallowing ... [https://www.nature.com/articles/s41528-023-00286-9]
• 11. Conformable AlN Piezoelectric Sensors as a Non-invasive ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC8294609/]
• 12. A new flexible piezoelectric pressure sensor array for the ... [https://jps.biomedcentral.com/articles/10.1007/s12576-018-0601-0]
• 13. Automatic Detection of Chewing and Swallowing [https://www.mdpi.com/1424-8220/21/10/3378]
• 14. New highly efficient material turns motion into power [https://www.birmingham.ac.uk/news/2025/new-highly-efficient-material-turns-motion-into-power-without-toxic-lead]
• 15. Development of a System to Monitor Laryngeal Movement ... [https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0070850]
• 16. Correlation of Temporal Parameters of Laryngeal ... [https://link.springer.com/article/10.1007/s00455-020-10121-2]
• 17. A new flexible piezoelectric pressure sensor array for the ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC10717778/]
• 18. Eating Event Recognition Using Accelerometer, Gyroscope ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11154557/]
• 19. Validity and Reliability of a Noninvasive Device for ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11906385/]
• 20. Expert-level aspiration and penetration detection during ... [https://www.nature.com/articles/s41598-022-25618-z]
• 21. Dysphagia assessment based on photoacoustic imaging [https://pmc.ncbi.nlm.nih.gov/articles/PMC12546401/]
• 22. 6 Benefits of FEES Swallowing Assessment [https://keytospeakandswallow.com.au/fees-swallowing-assessment/]
• 23. Endoscopic and videofluoroscopic evaluations of ... [https://www.sciencedirect.com/science/article/pii/S1808869425000412]
• 24. Early application value of flexible laryngoscope swallowing ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11839982/]
• 25. Advancements and Challenges in Non-Invasive Sensor ... [https://www.mdpi.com/2306-5354/11/5/430]
• 26. A new flexible piezoelectric pressure sensor array for the noninvasive detection of laryngeal movement during swallowing - PubMed [https://pubmed.ncbi.nlm.nih.gov/29508285/]
• 27. Applications of piezoelectric-based sensors, actuators, and ... [https://www.sciencedirect.com/science/article/pii/S2666053925000219]
• 28. A comparison of piezoelectric-based inertial sensing and ... [https://www.sciencedirect.com/science/article/abs/pii/S2451847615300105]
• 29. Automatic Detection of Chewing and Swallowing - PMC [https://pmc.ncbi.nlm.nih.gov/articles/PMC8152027/]
• 30. Global Lead-Free Piezoelectric Ceramics Market Report ... [https://www.bccresearch.com/market-research/nanotechnology/lead-free-piezoelectric-ceramics-market-report.html?srsltid=AfmBOoqh3zyDogYO8gpN4uiGiz_lTFzo-6Qe_FrVY1VkiUjLatnXoAqi]
• 31. Advances in lead-free flexible piezoelectric materials for ... [https://www.sciencedirect.com/science/article/pii/S2542504825000156]
• 32. Next‐Generation Piezoelectric Materials in Wearable and ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12591194/]
• 33. A Comprehensive Review of Piezoelectric PVDF Polymer ... [https://www.mdpi.com/2072-666X/16/4/386]
• 34. Enhanced energy harvesting from NF-PVDF piezoelectric ... [https://link.springer.com/article/10.1007/s40243-025-00321-x]
• 35. Environmentally friendly lead-free piezoelectric ceramics ... [https://www.sciencedirect.com/science/article/pii/S2214509525009726]
• 36. BNT Lead-free Piezoelectric Ceramics Market [https://www.futuremarketinsights.com/reports/bnt-lead-free-piezoelectric-ceramics-market]
• 37. On-Skin Epidermal Electronics for Next-Generation Health ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12334786/]
• 38. Application of an artificial intelligence–assisted electronic ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12336426/]
• 39. Recent advancements in wearable sensors: integration with ... [https://pubs.rsc.org/en/content/articlehtml/2025/ra/d5ra00167f]
• 40. Flexible wearable medical devices: from material ... [https://www.sciencedirect.com/science/article/pii/S2090123225006514]
• 41. Automatic Detection of Swallowing Events by Acoustical ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC2886152/]
• 42. Biometric Vibration Signal Detection Devices for ... [https://www.mdpi.com/2624-6120/5/3/28]
• 43. Machine learning techniques for the estimation of viscosity ... [https://www.sciencedirect.com/science/article/abs/pii/S0924424723005435]
• 44. Design and Development of an Electronic Interface for ... [https://www.mdpi.com/2227-7080/13/7/270]
• 45. Swallow Detection Algorithm Based on Bioimpedance and ... [https://www.sciencedirect.com/science/article/pii/S1474667016320791]
• 46. BNC interface lets piezoelectric sensors hook up to data ... [https://www.eeworldonline.com/picotech-bnc-interface-lets-piezoelectric-sensors-hook-up-to-data-acquisition-gear/]
• 47. Automated detection and recognition system for chewable ... [https://www.nature.com/articles/s41598-024-57077-z]
• 48. A Seismic Data Acquisition System Based on Wireless ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC8272099/]
• 49. Segmentation and Characterization of Chewing Bouts by ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC5515693/]
• 50. Artificial intelligence in the diagnosis and management of ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12337716/]
• 51. A noninvasive swallowing measurement system using a combination... [https://link.springer.com/article/10.1007/s11517-016-1561-2]
• 52. A hybrid model for detecting motion artifacts in ballistocardiogram signals | BioMedical Engineering OnLine | Full Text [https://biomedical-engineering-online.biomedcentral.com/articles/10.1186/s12938-025-01426-0]
• 53. Quantitative Comparison of the Performance ... [https://www.frontiersin.org/journals/physiology/articles/10.3389/fphys.2019.01563/full]
• 54. Advances in Wearable Piezoelectric Sensors for Hazardous Workplace Environments - PMC [https://pmc.ncbi.nlm.nih.gov/articles/PMC10242536/]
• 55. Advancements and Challenges in Non-Invasive Sensor ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC11118896/]
• 56. Flexible piezoelectrics: integration of sensing, actuating ... [https://www.nature.com/articles/s41528-025-00432-5]
• 57. Piezoelectric materials for flexible and wearable electronics [https://www.sciencedirect.com/science/article/pii/S026412752100719X]
• 58. The Neural Correlates of Chewing Gum—A Neuroimaging ... [https://www.mdpi.com/2076-3425/15/6/657]
• 59. Piezoelectric Devices Market Size, Share, Industry Report ... [https://www.marketsandmarkets.com/Market-Reports/piezoelectric-devices-market-256019882.html]
• 60. Advances in wearable electronics for monitoring human ... [https://www.sciencedirect.com/science/article/pii/S0142961224003995]
• 61. Energy Harvesting PMIC | Advanced PCB Design Blog [https://resources.pcb.cadence.com/blog/2023-energy-harvesting-pmic]
• 62. Harvesting Energy Efficiently [https://www.arrow.com/en/research-and-events/articles/harvesting-energy-efficiently-with-energy-management-techniques]
• 63. sEMG-based automatic characterization of swallowed materials [https://biomedical-engineering-online.biomedcentral.com/articles/10.1186/s12938-024-01241-z]
• 64. Pilot Study: Magnetic Motion Analysis for Swallowing ... [https://www.mdpi.com/1424-8220/23/7/3594]
• 65. A Systematic Review of Sensor-Based Methods for ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC12114971/]
• 66. Diagnostic utility of videofluoroscopic swallowing study in ... [https://pubmed.ncbi.nlm.nih.gov/41002123/]
• 67. High-Resolution Manometry—Observations After 15 Years ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC7413906/]
• 68. Development of a flexible balloon sensor for ... [https://www.sciencedirect.com/science/article/abs/pii/S0263224125003951]
• 69. Current Technological Advances in Dysphagia Screening [https://pmc.ncbi.nlm.nih.gov/articles/PMC12089864/]
• 70. Swallow Detection with Acoustics and Accelerometric ... [https://www.mdpi.com/1660-4601/20/1/170]
• 71. A Sensor System for Automatic Detection of Food Intake ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC3366471/]
• 72. Swallow Detection with Acoustics and Accelerometric ... [https://pmc.ncbi.nlm.nih.gov/articles/PMC9819201/]
• 73. Multimodal deep ensemble classification system with ... [https://www.nature.com/articles/s41746-024-01417-w]
• 74. A Systematic Review of Sensor-Based Methods for ... [https://www.mdpi.com/1424-8220/25/10/2966]