Abstract
Machine learning (ML) has emerged as a vital tool for the diagnosis of Parkinson's Disease (PD). This study presents a comprehensive review on the applications of ML for computer-aided diagnosis (CAD) of PD. We conducted a comprehensive review by searching articles published from 2010 till 2024. The risk of bias is assessed using the PROBAST checklist. Case studies are also provided. This review includes 117 articles with six categories: neuroimaging data (20.5%); voice data (40.2%); handwriting data (12.0%); gait data (14.5%); EEG data (8.5%); and other data (4.3%). According to the PROBAST checklist, only 28 articles (23.9%) have a low risk of bias. A benchmark case study is conducted for five different data modalities. We also discuss current limitations and future directions of applying ML to the diagnosis of PD. This review reduces the gap between Artificial Intelligence (AI) and PD medical professionals and provides helpful information for future research.