Swallow Detection Algorithm Based on Bioimpedance and EMG Measurements

Abstract

Image-based swallowing assessment tools like videofluoroscopy and endoscopy allow experts manual investigation of a few individual swallows. However, these tools are expensive and can only be used by clinicians. Systems which utilize easily attachable, inexpensive and non-invasive sensors at the throat could be a real progress for diagnosis and therapy. This contribution investigates the use of a combined electromyography (EMG) and bioimpedance (BI) measurement at the throat to automatically detect swallowing events. The absolute value of the measured BI completely describes the swallowing process, i.e. the closure of the larynx. There is a typical reproducible drop in BI during a swallow. The muscle activity needed for the laryngeal movement during a swallow is measured using EMG. The presented algorithm involves a valley detection in order to perform a segmentation of the BI signal. Additionally, only BI valleys that coincide with EMG activity are selected for feature extraction. In the second part of the algorithm, extracted features of the BI and integrated EMG are fed into a support vector machine (SVM) which is able to separate BI valleys related to swallowing events from valleys which are not caused by swallowing. The detection algorithm has been tested on data from nine healthy subjects. The data set contained 1370 swallows of different bolus sizes and consistency and was effected by other movements and speech. The combined BI/EMG segmentation detected 99.3% of all swallowing events. The subsequently applied classifier showed a sensitivity of 96.1% and a specificity of 97.1% for the test data.