Bradykinesia Detection in Parkinson’s Disease Using Machine Learning Technique

Abstract

Bradykinesia is the most side effect of Parkinson’s Diease (PD), and poses significant challenges in motor function, complicating accurate and timely diagnosis. This survey investigates the implementation of machine learning (ML) techniques in detecting and quantifying bradykinesia among PD patients. This survey evaluated the performance of various machine learning models, including Variational Autoencoders (VAEs), ROCKET, and InceptionTime, using the GENEActiv dataset to detect bradykinesia. The survey delves into key features extracted for ML models—such as movement speed, rhythm, and amplitude—highlighting their relevance in enhancing diagnostic precision. Micheal J. Fox Foundation (MJFF) Levodopa Response dataset is used as an input for the previously mentioned Machine Learning Models. All participants wore a sensor device (GeneActiv) on the wrist of their most affected limb. Our findings also highlight the potential of the InceptionTime and ROCKET models, 0.673 and 0.567 of mean average precision and balanced accuracy respectively for the InceptionTime model. The ROCKET model achieved 0.727 of mean average precision.