Electrophysiological and Mechanical Approaches to the Swallowing Analysis

Abstract

Alterations of the swallowing process at neurological and muscular levels can cause dysphagia. Its instrumental diagnosis is carried out by videofluoroscopy, which is invasive, expensive, and sometimes unavailable. Inexpensive and non-invasive alternatives have been proposed in several studies considering different biosignals including surface electromyography and accelerometry. Automatic approaches have been focused on unimodal analysis, disregarding the high complexity of the process. In this way, we performed a multimodal analysis of the swallowing process in 30 healthy individuals and 30 dysphagic patients, using surface electromyography and accelerometry-based cervical auscultation. Features in time, frequency, and time-frequency domains were extracted and selected for classification purposes. A support vector machine was used as a classifier, and hyperparameters were optimized by a grid search. The proposed scheme achieved an accuracy of 0.92 ± 0.07 in the best case. The performed method achieved promising results in terms of classification of dysphagic and healthy individuals, which could contribute to the development of non-invasive techniques of dysphagia screening.