Deep Learning for Classification of Normal Swallows in Adults

Swallowing sound index analysis using electronic stethoscope and artificial intelligence for patients with Parkinson's disease

BACKGROUND AND PURPOSE

Several noninvasive tools assess swallowing disorders, including electronic stethoscope artificial intelligence (AI) analysis for remote diagnosis, with the potential for telemedicine. This study investigated the swallowing sound index in patients with Parkinson's disease (PD).

METHODS

This single-arm, open-label trial assessed the impact of cervical percutaneous interferential current stimulation on swallowing in patients with PD classified as Hoehn-Yahr stages 2-4. Stimulation was conducted for 8 weeks. Baseline data were used to examine the link between the swallowing sound index and indicators such as videofluoroscopy (VF). Furthermore, we examined changes in the swallowing sound index after the intervention.

RESULTS

Twenty-five patients were included. The swallowing sound index in patients with PD was higher than that in those with amyotrophic lateral sclerosis but considerably lower than that in healthy controls. The number of patients with normal EAT-10 scores positively correlated with the swallowing sound index, whereas elevated C-reactive protein levels were negatively correlated with the swallowing sound index. However, the index displayed no correlation with other indicators, including the VF results. Despite the intervention, the index remained unchanged throughout the study.

CONCLUSION

In patients with PD, a decrease in the swallowing sound index suggests a potential association between swallowing disorders and the risk of aspiration pneumonia.

TRIAL REGISTRATION NUMBER

jRCTs062220013.

Swallowing sound evaluation using an electronic stethoscope and artificial intelligence analysis for patients with amyotrophic lateral sclerosis

Background and purpose

Non-invasive, simple, and repetitive swallowing evaluation is required to prevent aspiration pneumonia in neurological care. We investigated the usefulness of swallowing sound evaluation in patients with amyotrophic lateral sclerosis (ALS) using our new electronic stethoscope artificial intelligence (AI) analysis tool.

Methods

We studied patients with ALS who provided written informed consent. We used an electronic stethoscope, placed a Bluetooth-enabled electronic stethoscope on the upper end of the sternum, performed a 3-mL water swallow three times, and remotely identified the intermittent sound components of the water flow caused at that time by AI, with the maximum value as the swallowing sound index. We examined the correlation between the swallowing sound index and patient background, including swallowing-related parameters.

Results

We evaluated 24 patients with ALS (age 64.0 ± 11.8 years, 13 women, median duration of illness 17.5 months). The median ALS Functional Rating Scale-Revised (ALSFRS-R) score was 41 (minimum 18, maximum 47). In all cases, the mean swallowing sound index was 0.209 ± 0.088. A multivariate analysis showed that a decrease in the swallowing sound index was significantly associated with a low ALSFRS-R score, an ALSFRS-R bulbar symptom score, % vital capacity, tongue pressure, a Mann Assessment of Swallowing Ability (MASA) score, and a MASA pharyngeal phase-related score.

Conclusion

Swallowing sound evaluation using an electronic stethoscope AI analysis showed a correlation with existing indicators in swallowing evaluation in ALS and suggested its usefulness as a new method. This is expected to be a useful examination method in home and remote medical care.

Toward a robust swallowing detection for an implantable active artificial larynx: a survey

Total laryngectomy consists in the removal of the larynx and is intended as a curative treatment for laryngeal cancer, but it leaves the patient with no possibility to breathe, talk, and swallow normally anymore. A tracheostomy is created to restore breathing through the throat, but the aero-digestive tracts are permanently separated and the air no longer passes through the nasal tracts, which allowed filtration, warming, humidification, olfaction, and acceleration of the air for better tissue oxygenation. As for phonation restoration, various techniques allow the patient to talk again. The main one consists of a tracheo-esophageal valve prosthesis that makes the air passes from the esophagus to the pharynx, and makes the air vibrate to allow speech through articulation. Finally, swallowing is possible through the original tract as it is now isolated from the trachea. Yet, many methods exist to detect and assess a swallowing, but none is intended as a definitive restoration technique of the natural airway, which would permanently close the tracheostomy and avoid its adverse effects. In addition, these methods are non-invasive and lack detection accuracy. The feasibility of an effective early detection of swallowing would allow to further develop an implantable active artificial larynx and therefore restore the aero-digestive tracts. A previous attempt has been made on an artificial larynx implanted in 2012, but no active detection was included and the system was completely mechanic. This led to residues in the airway because of the imperfect sealing of the mechanism. An active swallowing detection coupled with indwelling measurements would thus likely add a significant reliability on such a system as it would allow to actively close an artificial larynx. So, after a brief explanation of the swallowing mechanism, this survey intends to first provide a detailed consideration of the anatomical region involved in swallowing, with a detection perspective. Second, the swallowing mechanism following total laryngectomy surgery is detailed. Third, the current non-invasive swallowing detection technique and their limitations are discussed. Finally, the previous points are explored with regard to the inherent requirements for the feasibility of an effective swallowing detection for an artificial larynx. Graphical Abstract.

Autonomous Swallow Segment Extraction Using Deep Learning in Neck-Sensor Vibratory Signals From Patients With Dysphagia

Dysphagia occurs secondary to a variety of underlying etiologies and can contribute to increased risk of adverse events such as aspiration pneumonia and premature mortality. Dysphagia is primarily diagnosed and characterized by instrumental swallowing exams such as videofluoroscopic swallowing studies. videofluoroscopic swallowing studies involve the inspection of a series of radiographic images for signs of swallowing dysfunction. Though effective, videofluoroscopic swallowing studies are only available in certain clinical settings and are not always desirable or feasible for certain patients. Because of the limitations of current instrumental swallow exams, research studies have explored the use of acceleration signals collected from neck sensors and demonstrated their potential in providing comparable radiation-free diagnostic value as videofluoroscopic swallowing studies. In this study, we used a hybrid deep convolutional recurrent neural network that can perform multi-level feature extraction (localized and across time) to annotate swallow segments automatically via multi-channel swallowing acceleration signals. In total, we used signals and videofluoroscopic swallowing study images of 3144 swallows from 248 patients with suspected dysphagia. Compared to other deep network variants, our network was superior at detecting swallow segments with an average area under the receiver operating characteristic curve value of 0.82 (95% confidence interval: 0.807-0.841), and was in agreement with up to 90% of the gold standard-labeled segments.

Research on a real-time dynamic monitoring method for silent aspiration after stroke based on semisupervised deep learning: A protocol study

Objective

This study aims to establish a real-time dynamic monitoring system for silent aspiration (SA) to provide evidence for the early diagnosis of and precise intervention for SA after stroke.

Methods

Multisource signals, including sound, nasal airflow, electromyographic, pressure and acceleration signals, will be obtained by multisource sensors during swallowing events. The extracted signals will be labeled according to videofluoroscopic swallowing studies (VFSSs) and input into a special dataset. Then, a real-time dynamic monitoring model for SA will be built and trained based on semisupervised deep learning. Model optimization will be performed based on the mapping relationship between multisource signals and insula-centered cerebral cortex-brainstem functional connectivity through resting-state functional magnetic resonance imaging. Finally, a real-time dynamic monitoring system for SA will be established, of which the sensitivity and specificity will be improved by clinical application.

Results

Multisource signals will be stably extracted by multisource sensors. Data from a total of 3200 swallows will be obtained from patients with SA, including 1200 labeled swallows from the nonaspiration category from VFSSs and 2000 unlabeled swallows. A significant difference in the multisource signals is expected to be found between the SA and nonaspiration groups. The features of labeled and pseudolabeled multisource signals will be extracted through semisupervised deep learning to establish a dynamic monitoring model for SA. Moreover, strong correlations are expected to be found between the Granger causality analysis (GCA) value (from the left middle frontal gyrus to the right anterior insula) and the laryngeal rise time (LRT). Finally, a dynamic monitoring system will be established based on the former model, by which SA can be identified precisely.

Conclusion

The study will establish a real-time dynamic monitoring system for SA with high sensitivity, specificity, accuracy and F1 score.

Using an Automated Speech Recognition Approach to Differentiate Between Normal and Aspirating Swallowing Sounds Recorded from Digital Cervical Auscultation in Children

Use of machine learning to accurately detect aspirating swallowing sounds in children is an evolving field. Previously reported classifiers for the detection of aspirating swallowing sounds in children have reported sensitivities between 79 and 89%. This study aimed to investigate the accuracy of using an automatic speaker recognition approach to differentiate between normal and aspirating swallowing sounds recorded from digital cervical auscultation in children. We analysed 106 normal swallows from 23 healthy children (median 13 months; 52.1% male) and 18 aspirating swallows from 18 children (median 10.5 months; 61.1% male) who underwent concurrent videofluoroscopic swallow studies with digital cervical auscultation. All swallowing sounds were on thin fluids. A support vector machine classifier with a polynomial kernel was trained on feature vectors that comprised the mean and standard deviation of spectral subband centroids extracted from each swallowing sound in the training set. The trained support vector machine was then used to classify swallowing sounds in the test set. We found high accuracy in the differentiation of aspirating and normal swallowing sounds with 98% overall accuracy. Sensitivity for the detection of aspiration and normal swallowing sounds were 89% and 100%, respectively. There were consistent differences in time, power spectral density and spectral subband centroid features between aspirating and normal swallowing sounds in children. This study provides preliminary research evidence that aspirating and normal swallowing sounds in children can be differentiated accurately using machine learning techniques.

A Preliminary Investigation of Similarities of High Resolution Cervical Auscultation Signals Between Thin Liquid Barium and Water Swallows

Dysphagia, commonly referred to as abnormal swallowing, affects millions of people annually. If not diagnosed expeditiously, dysphagia can lead to more severe complications, such as pneumonia, nutritional deficiency, and dehydration. Bedside screening is the first step of dysphagia characterization and is usually based on pass/fail tests in which a nurse observes the patient performing water swallows to look for dysphagia overt signs such as coughing. Though quick and convenient, bedside screening only provides low-level judgment of impairment, lacks standardization, and suffers from subjectivity. Recently, high resolution cervical auscultation (HRCA) has been investigated as a less expensive and non-invasive method to diagnose dysphagia. It has shown strong preliminary evidence of its effectiveness in penetration-aspiration detection as well as multiple swallow kinematics. HRCA signals have traditionally been collected and investigated in conjunction with videofluoroscopy exams which are performed using barium boluses including thin liquid. An HRCA-based bedside screening is highly desirable to expedite the initial dysphagia diagnosis and overcome all the drawbacks of the current pass/fail screening tests. However, all research conducted for using HRCA in dysphagia is based on thin liquid barium boluses and thus not guaranteed to provide valid results for water boluses used in bedside screening. If HRCA signals show no significant differences between water and thin liquid barium boluses, then the same algorithms developed on thin liquid barium boluses used in diagnostic imaging studies, it can be then directly used with water boluses. This study investigates the similarities and differences between HRCA signals from thin liquid barium swallows compared to those signals from water swallows. Multiple features from the time, frequency, time-frequency, and information-theoretic domain were extracted from each type of swallow and a group of linear mixed models was tested to determine the significance of differences. Machine learning classifiers were fit to the data as well to determine if the swallowed material (thin liquid barium or water) can be correctly predicted from an unlabeled set of HRCA signals. The results demonstrated that there is no systematic difference between the HRCA signals of thin liquid barium swallows and water swallows. While no systematic difference was discovered, the evidence of complete conformity between HRCA signals of both materials was inconclusive. These results must be validated further to confirm conformity between the HRCA signals of thin liquid barium swallows and water swallows.

Characterizing Effortful Swallows from Healthy Community Dwelling Adults Across the Lifespan Using High-Resolution Cervical Auscultation Signals and MBSImP Scores: A Preliminary Study

There is growing enthusiasm to develop inexpensive, non-invasive, and portable methods that accurately assess swallowing and provide biofeedback during dysphagia treatment. High-resolution cervical auscultation (HRCA), which uses acoustic and vibratory signals from non-invasive sensors attached to the anterior laryngeal framework during swallowing, is a novel method for quantifying swallowing physiology via advanced signal processing and machine learning techniques. HRCA has demonstrated potential as a dysphagia screening method and diagnostic adjunct to VFSSs by determining swallowing safety, annotating swallow kinematic events, and classifying swallows between healthy participants and patients with a high degree of accuracy. However, its feasibility as a non-invasive biofeedback system has not been explored. This study investigated 1. Whether HRCA can accurately differentiate between non-effortful and effortful swallows; 2. Whether differences exist in Modified Barium Swallow Impairment Profile (MBSImP) scores (#9, #11, #14) between non-effortful and effortful swallows. We hypothesized that HRCA would accurately classify non-effortful and effortful swallows and that differences in MBSImP scores would exist between the types of swallows. We analyzed 247 thin liquid 3 mL command swallows (71 effortful) to minimize variation from 36 healthy adults who underwent standardized VFSSs with concurrent HRCA. Results revealed differences (p < 0.05) in 9 HRCA signal features between non-effortful and effortful swallows. Using HRCA signal features as input, decision trees classified swallows with 76% accuracy, 76% sensitivity, and 77% specificity. There were no differences in MBSImP component scores between non-effortful and effortful swallows. While preliminary in nature, this study demonstrates the feasibility/promise of HRCA as a biofeedback method for dysphagia treatment.

Characterizing Swallows From People With Neurodegenerative Diseases Using High-Resolution Cervical Auscultation Signals and Temporal and Spatial Swallow Kinematic Measurements

Purpose The prevalence of dysphagia in patients with neurodegenerative diseases (ND) is alarmingly high and frequently results in morbidity and accelerated mortality due to subsequent adverse events (e.g., aspiration pneumonia). Swallowing in patients with ND should be continuously monitored due to the progressive disease nature. Access to instrumental swallow evaluations can be challenging, and limited studies have quantified changes in temporal/spatial swallow kinematic measures in patients with ND. High-resolution cervical auscultation (HRCA), a dysphagia screening method, has accurately differentiated between safe and unsafe swallows, identified swallow kinematic events (e.g., laryngeal vestibule closure [LVC]), and classified swallows between healthy adults and patients with ND. This study aimed to (a) compare temporal/spatial swallow kinematic measures between patients with ND and healthy adults and (b) investigate HRCA's ability to annotate swallow kinematic events in patients with ND. We hypothesized there would be significant differences in temporal/spatial swallow measurements between groups and that HRCA would accurately annotate swallow kinematic events in patients with ND. Method Participants underwent videofluoroscopic swallowing studies with concurrent HRCA. We used linear mixed models to compare temporal/spatial swallow measurements (n = 170 ND patient swallows, n = 171 healthy adult swallows) and deep learning machine-learning algorithms to annotate specific temporal and spatial kinematic events in swallows from patients with ND. Results Differences (p < .05) were found between groups for several temporal and spatial swallow kinematic measures. HRCA signal features were used as input to machine-learning algorithms and annotated upper esophageal sphincter (UES) opening, UES closure, LVC, laryngeal vestibule reopening, and hyoid bone displacement with 66.25%, 85%, 68.18%, 70.45%, and 44.6% accuracy, respectively, compared to human judges' measurements. Conclusion This study demonstrates HRCA's potential in characterizing swallow function in patients with ND and other patient populations.

Establishing Reference Values for Temporal Kinematic Swallow Events Across the Lifespan in Healthy Community Dwelling Adults Using High-Resolution Cervical Auscultation

Few research studies have investigated temporal kinematic swallow events in healthy adults to establish normative reference values. Determining cutoffs for normal and disordered swallowing is vital for differentially diagnosing presbyphagia, variants of normal swallowing, and dysphagia; and for ensuring that different swallowing research laboratories produce consistent results in common measurements from different samples within the same population. High-resolution cervical auscultation (HRCA), a sensor-based dysphagia screening method, has accurately annotated temporal kinematic swallow events in patients with dysphagia, but hasn't been used to annotate temporal kinematic swallow events in healthy adults to establish dysphagia screening cutoffs. This study aimed to determine: (1) Reference values for temporal kinematic swallow events, (2) Whether HRCA can annotate temporal kinematic swallow events in healthy adults. We hypothesized (1) Our reference values would align with a prior study; (2) HRCA would detect temporal kinematic swallow events as accurately as human judges. Trained judges completed temporal kinematic measurements on 659 swallows (N = 70 adults). Swallow reaction time and LVC duration weren't different (p > 0.05) from a previously published historical cohort (114 swallows, N = 38 adults), while other temporal kinematic measurements were different (p < 0.05), suggesting a need for further standardization to feasibly pool data analyses across laboratories. HRCA signal features were used as input to machine learning algorithms and annotated UES opening (69.96% accuracy), UES closure (64.52% accuracy), LVC (52.56% accuracy), and LV re-opening (69.97% accuracy); providing preliminary evidence that HRCA can noninvasively and accurately annotate temporal kinematic measurements in healthy adults to determine dysphagia screening cutoffs.

Anterior-posterior distension of maximal upper esophageal sphincter opening is correlated with high-resolution cervical auscultation signal features

Objective. Adequate upper esophageal sphincter (UES) opening is essential during swallowing to enable clearance of material into the digestive system, and videofluoroscopy (VF) is the most commonly deployed instrumental examination for assessment of UES opening. High-resolution cervical auscultation (HRCA) has been shown to be an effective, portable and cost-efficient screening tool for dysphagia with strong capabilities in non-invasively and accurately approximating manual measurements of VF images. In this study, we aimed to examine whether the HRCA signals are correlated to the manually measured anterior-posterior (AP) distension of maximal UES opening from VF recordings, under the hypothesis that they would be strongly associated.Approach. We developed a standardized method to spatially measure the AP distension of maximal UES opening in 203 swallows VF recording from 27 patients referred for VF due to suspected dysphagia. Statistical analysis was conducted to compare the manually measured AP distension of maximal UES opening from lateral plane VF images and features extracted from two sets of HRCA signal segments: whole swallow segments and segments excluding all events other than the duration of UES is opening.Main results. HRCA signal features were significantly associated with the normalized AP distension of the maximal UES opening in the longer whole swallowing segments and the association became much stronger when analysis was performed solely during the duration of UES opening.Significance. This preliminary feasibility study demonstrated the potential value of HRCA signals features in approximating the objective measurements of maximal UES AP distension and paves the way of developing HRCA to non-invasively and accurately predict human spatial measurement of VF kinematic events.

How Closely do Machine Ratings of Duration of UES Opening During Videofluoroscopy Approximate Clinician Ratings Using Temporal Kinematic Analyses and the MBSImP?

Clinicians evaluate swallow kinematic events by analyzing videofluoroscopy (VF) images for dysphagia management. The duration of upper esophageal sphincter opening (DUESO) is one important temporal swallow event, because reduced DUESO can result in pharyngeal residue and penetration/aspiration. VF is frequently used for evaluating swallowing but exposes patients to radiation and is not always feasible/readily available. High resolution cervical auscultation (HRCA) is a non-invasive, sensor-based dysphagia screening method that uses signal processing and machine learning to characterize swallowing. We investigated HRCA's ability to annotate DUESO and predict Modified Barium Swallow Impairment Profile (MBSImP) scores (component #14). We hypothesized that HRCA and machine learning techniques would detect DUESO with similar accuracy as human judges. Trained judges completed temporal kinematic measurements of DUESO on 719 swallows (116 patients) and 50 swallows (15 age-matched healthy adults). An MBSImP certified clinician completed MBSImP ratings on 100 swallows. A multi-layer convolutional recurrent neural network (CRNN) using HRCA signal features for input was used to detect DUESO. Generalized estimating equations models were used to determine statistically significant HRCA signal features for predicting DUESO MBSImP scores. A support vector machine (SVM) classifier and a leave-one-out procedure was used to predict DUESO MBSImP scores. The CRNN detected UES opening within a 3-frame tolerance for 82.6% of patient and 86% of healthy swallows and UES closure for 72.3% of patient and 64% of healthy swallows. The SVM classifier predicted DUESO MBSImP scores with 85.7% accuracy. This study provides evidence of HRCA's feasibility in detecting DUESO without VF images.

A Preliminary Investigation of Whether HRCA Signals Can Differentiate Between Swallows from Healthy People and Swallows from People with Neurodegenerative Diseases

High-resolution cervical auscultation (HRCA) is an emerging method for non-invasively assessing swallowing by using acoustic signals from a contact microphone, vibratory signals from an accelerometer, and advanced signal processing and machine learning techniques. HRCA has differentiated between safe and unsafe swallows, predicted components of the Modified Barium Swallow Impairment Profile, and predicted kinematic events of swallowing such as hyoid bone displacement, laryngeal vestibular closure, and upper esophageal sphincter opening with a high degree of accuracy. However, HRCA has not been used to characterize swallow function in specific patient populations. This study investigated the ability of HRCA to differentiate between swallows from healthy people and people with neurodegenerative diseases. We hypothesized that HRCA would differentiate between swallows from healthy people and people with neurodegenerative diseases with a high degree of accuracy. We analyzed 170 swallows from 20 patients with neurodegenerative diseases and 170 swallows from 51 healthy age-matched adults who underwent concurrent video fluoroscopy with non-invasive neck sensors. We used a linear mixed model and several supervised machine learning classifiers that use HRCA signal features and a leave-one-out procedure to differentiate between swallows. Twenty-two HRCA signal features were statistically significant (p < 0.05) for predicting whether swallows were from healthy people or from patients with neurodegenerative diseases. Using the HRCA signal features alone, logistic regression and decision trees classified swallows between the two groups with 99% accuracy, 100% sensitivity, and 99% specificity. This provides preliminary research evidence that HRCA can differentiate swallow function between healthy and patient populations.

High-Resolution Cervical Auscultation and Data Science: New Tools to Address an Old Problem

High-resolution cervical auscultation (HRCA) is an evolving clinical method for noninvasive screening of dysphagia that relies on data science, machine learning, and wearable sensors to investigate the characteristics of disordered swallowing function in people with dysphagia. HRCA has shown promising results in categorizing normal and disordered swallowing (i.e., screening) independent of human input, identifying a variety of swallowing physiological events as accurately as trained human judges. The system has been developed through a collaboration of data scientists, computer-electrical engineers, and speech-language pathologists. Its potential to automate dysphagia screening and contribute to evaluation lies in its noninvasive nature (wearable electronic sensors) and its growing ability to accurately replicate human judgments of swallowing data typically formed on the basis of videofluoroscopic imaging data. Potential contributions of HRCA when videofluoroscopic swallowing study may be unavailable, undesired, or not feasible for many patients in various settings are discussed, along with the development and capabilities of HRCA. The use of technological advances and wearable devices can extend the dysphagia clinician's reach and reinforce top-of-license practice for patients with swallowing disorders.

Tracking Hyoid Bone Displacement During Swallowing Without Videofluoroscopy Using Machine Learning of Vibratory Signals

Identifying physiological impairments of swallowing is essential for determining accurate diagnosis and appropriate treatment for patients with dysphagia. The hyoid bone is an anatomical landmark commonly monitored during analysis of videofluoroscopic swallow studies (VFSSs). Its displacement is predictive of penetration/aspiration and is associated with other swallow kinematic events. However, VFSSs are not always readily available/feasible and expose patients to radiation. High-resolution cervical auscultation (HRCA), which uses acoustic and vibratory signals from a microphone and tri-axial accelerometer, is under investigation as a non-invasive dysphagia screening method and potential adjunct to VFSS when it is unavailable or not feasible. We investigated the ability of HRCA to independently track hyoid bone displacement during swallowing with similar accuracy to VFSS, by analyzing vibratory signals from a tri-axial accelerometer using machine learning techniques. We hypothesized HRCA would track hyoid bone displacement with a high degree of accuracy compared to humans. Trained judges completed frame-by-frame analysis of hyoid bone displacement on 400 swallows from 114 patients and 48 swallows from 16 age-matched healthy adults. Extracted features from vibratory signals were used to train the predictive algorithm to generate a bounding box surrounding the hyoid body on each frame. A metric of relative overlapped percentage (ROP) compared human and machine ratings. The mean ROP for all swallows analyzed was 50.75%, indicating > 50% of the bounding box containing the hyoid bone was accurately predicted in every frame. This provides evidence of the feasibility of accurate, automated hyoid bone displacement tracking using HRCA signals without use of VFSS images.

Classifying Dysphagic Swallowing Sounds with Support Vector Machines

Swallowing sounds from cervical auscultation include information related to the swallowing function. Several studies have been conducted on the screening tests of dysphagia. The literature shows a significant difference between the characteristics of swallowing sounds obtained from different subjects (e.g., healthy and dysphagic subjects; young and old adults). These studies demonstrate the usefulness of swallowing sounds during dysphagic screening. However, the degree of classification for dysphagia based on swallowing sounds has not been thoroughly studied. In this study, we investigate the use of machine learning for classifying swallowing sounds into various types, such as normal swallowing or mild, moderate, and severe dysphagia. In particular, swallowing sounds were recorded from patients with dysphagia. Support vector machines (SVMs) were trained using some features extracted from the obtained swallowing sounds. Moreover, the accuracy of the classification of swallowing sounds using the trained SVMs was evaluated via cross-validation techniques. In the two-class scenario, wherein the swallowing sounds were divided into two categories (viz. normal and dysphagic subjects), the maximum F-measure was 78.9%. In the four-class scenario, where the swallowing sounds were divided into four categories (viz. normal subject, and mild, moderate, and severe dysphagic subjects), the F-measure values for the classes were 65.6%, 53.1%, 51.1%, and 37.1%, respectively.

Automatic Detection of the Pharyngeal Phase in Raw Videos for the Videofluoroscopic Swallowing Study Using Efficient Data Collection and 3D Convolutional Networks †

Videofluoroscopic swallowing study (VFSS) is a standard diagnostic tool for dysphagia. To detect the presence of aspiration during a swallow, a manual search is commonly used to mark the time intervals of the pharyngeal phase on the corresponding VFSS image. In this study, we present a novel approach that uses 3D convolutional networks to detect the pharyngeal phase in raw VFSS videos without manual annotations. For efficient collection of training data, we propose a cascade framework which no longer requires time intervals of the swallowing process nor the manual marking of anatomical positions for detection. For video classification, we applied the inflated 3D convolutional network (I3D), one of the state-of-the-art network for action classification, as a baseline architecture. We also present a modified 3D convolutional network architecture that is derived from the baseline I3D architecture. The classification and detection performance of these two architectures were evaluated for comparison. The experimental results show that the proposed model outperformed the baseline I3D model in the condition where both models are trained with random weights. We conclude that the proposed method greatly reduces the examination time of the VFSS images with a low miss rate.