A Wearable System for Monitoring Neurological Disorder Events with Multi-Class Classification Model in Daily Life.

Abstract

Dysphagia and dysarthria are the prominent sequelae of neurological disorders. Treatment and rehabilitation of these impairments necessitate continuously monitoring symptoms related to swallowing and speaking. However, current medical technologies require large and diverse equipment to record these symptoms, which are predominantly limited to clinical environments. In this study, we propose an innovative wearable system for distinguishing neurological disorder events using a mechano-acoustic (MA) sensor and multi-class ensemble classification model. The MA sensor exhibits a high sensitivity to neck vibration without any interference from ambient sounds. A multi-class classification model was also developed to discern the symptoms from the recorded signals accurately. The proposed classification model is an ensemble neural network trained on waveforms and mel spectrograms. As a result, we achieve a high classification accuracy of 91.94%, surpassing the performance of previous single neural networks.