Deep Learning Models for Parkinson’s Disease Severity Evaluation

This chapter conducts a comprehensive exploration into the application of Machine Learning (ML) and Deep Learning (DL) algorithms for the diagnosis and assessment of Parkinson’s Disease (PD) severity through remote and continuous monitoring. The chapter is divided into two sections, outlining the methodology and results of the study. In the first section, an A-WEAR bracelet is employed for the objective assessment of tremor and bradykinesia in PD subjects and healthy older adults. To distinguish patients from healthy controls, temporal and spectral features are extracted, with non-linear temporal and spectral features demonstrating substantial differences. Both supervised and unsupervised ML classifiers yield favorable results. In the second section, a resampling technique is adopted to address the issue of an unbalanced dataset. Time and frequency-based features are extracted, and the signals are analyzed using the CatBoost classifier. This approach yields an impressive accuracy rate of 96%. The findings of this study substantially contribute to the development of a reliable and accurate framework for assessing PD severity.