Dysphagia diagnosis system with integrated speech analysis from throat vibration

Abstract

Dysphagia is a widespread swallowing disorder. One of the most widely used clinical assessments is the bedside swallowing disorder screening, including the water swallow test. This method can only diagnose the disease according to the choking situation of the liquid, and the aspiration may appear during the drinking water test. In this paper, an automatic diagnostic system is designed according to the patient’s performance in pronunciation. To diagnose dysphagia, we treat it as a binary classification task. We use a throat vibrator to collect the vibration signals of the subjects during pronunciation for clear speech signals. Feature engineering techniques, including feature extraction, data balancing and dimensionality reduction are implemented to extract the features for classification. Three popular classifiers are integrated to make the final predictions. Experimental results show that the integrated classifier performs better with accuracy = 72.09%, sensitivity = 63.64% and specificity = 80.95% respectively. Different from the traditional clinical ways, a new machine learning based swallowing disorders detection system is proposed with vibration signals of throat acquired for high-quality speech analysis. It may help to realize a portal and low-cost telemedicine device for dysphagia diagnosis at home.