A Mobile Swallow Monitoring System Based On Throat-worn Device

Swallow,as one of the most complex somatic reflexes,is a mirror of our daily ingestion and physical health.Chronic disease,for example,Parkinson disease and stroke,may cause swallow difficulty(also called dysphagia).Dysphagia influences the normal life and daily food consumption,and it may even causes pulmonary infection.The proportion of people with dysphagia in the elderly is rather high.With the heightened awareness of health care and the development of semi-conductor technology,applying mobile devices to swallow monitoring can be a rising interdisciplinary.In this paper,we propose a mobile swallow monitoring system.The system utilize a flexible throat-worn device to acquire photoplethysmogram(PPG),acceleration(ACC)and angular velocity.We are the first to utilize PPG signal characteristic to conduct swallow related research.Inertial sensor can also catch the movement of laryngeal cartilage during swallowing.The data is then transmitted to an Android app via Bluetooth to be displayed,monitored and recorded.Based on the wearable device,we also propose a swallow detection algorithm and a swallow function assessment algorithm.The swallow detection algorithm,aiming to judge whether a person is swallowing or not,utilizes support vector machine(SVM)to train ACC and PPG model with a small number of ACC and PPG features separately.And then,it fuses the output of the two models by logistic regression(LR).The SVM models and the LR coefficients are then used for swallow region segmentation in the swallow function assessment algorithm.The latter algorithm aims to distinguish between normal and abnorlal swallow.It utilizes F-score to conduct feature selection from a sets of features,and the final result is given by an SVM model.Two experiments are conducted to verify the algorithms.In the swallow detection experiment,swallowing and non-swallowing waveforms are recorded from 20 health subjects.The algorithm achieves 100.0%sensitivity,90.5%precision as well as 60.0%specificity,higher than the PPG based detection and the ACC based detection.In the swallow function assessment experiment,80 people in inpatient units are asked to conduct water swallow test,and normal and abnormal swallow waveforms are recorded.The algorithm gets 69.231%specificity for normal swallow detection and 70.588%sensitivity for abnormal swallow detection.What’s more,the algorithm achieves 87.500%specificity and 68.182%precision in people aged 64 and below.By adjusting the mode’s parameters,the algorithm can achieve higher specificity.In the future,the system can be used for daily swallow monitoring and early dysphagia screening in non-hospital environment.