A Graded Assessment System for Parkinson's Upper-Limb Bradykinesia Based on a Temporal Convolutional Network Model

Parkinson's disease (PD) is a kind of neurological diseases. As one of its most common symptoms, bradykinesia mainly manifests as the slowing of fine movements of the upper limbs, and it seriously affects the quality of life. In this article, a temporal convolutional network (TCN) model-based Parkinson's bradykinesia grading evaluation system is proposed. Firstly, a wearable device based on inertial sensors and low-energy Bluetooth was developed to collect kinematic information, 66 subjects from the Peking Union Medical College Hospital participated in the data collection work, and a dataset with four degrees of bradykinesia (Normal, Mild, Moderate, and Severe) was built. Afterward, a TCN model containing six residual block (RB) layers was designed for the evaluation of bradykinesia grades. The method showed better performance than the methods commonly used for bradykinesia detection in the comparison experiments. It provides a feasible method of PD bradykinesia grades assessment for assisting clinical diagnosis.