Comparative Evaluation of Deep Neural Models for Parkinson’s Disease Monitoring Using Multimodal Wearable Sensor Data

Abstract

Parkinson’s disease (PD) is a progressive neurological disorder for which continuous observation of motor and non-motor manifestations can support clinical assessment and long-term disease management. This study examines multimodal wearable sensor data and deep neural models for PD-related pattern analysis in a home-oriented monitoring setting. The framework integrates accelerometer, electromyography (EMG), gyroscope, and acoustic measurements with edge-level processing using a Raspberry Pi platform and wireless communication. Sensor observations collected during structured daily activities were used to compare convolutional, recurrent, and autoencoder-based deep-learning approaches reported in the source study. The primary experimental cohort reported in the methodology consisted of 50 participants, including 35 individuals with PD and 15 control participants. The source manuscript reports performance values between 97.8% and 99.8%; however, the final manuscript must explicitly define the metric and validation protocol supporting these values. The findings demonstrate the potential of multimodal sensing and deep learning for continuous PD monitoring while emphasizing that algorithmic outputs are decision-support signals rather than substitutes for clinical diagnosis.