Classification of Parkinson's Disease Using Machine Learning with MoCA Response Dynamics

Neurodegenerative diseases (NDs), including Parkinson's and Alzheimer's disease, pose a significant challenge to global health, and early detection tools are crucial for effective intervention. The adaptation of online screening forms and machine learning methods can lead to better and wider diagnosis, potentially altering the progression of NDs. Therefore, this study examines the diagnostic efficiency of machine learning models using Montreal Cognitive Assessment test results (MoCA) to classify scores of people with Parkinson's disease (PD) and healthy subjects. For data analysis, we implemented both rule-based modeling using rough set theory (RST) and classic machine learning (ML) techniques such as logistic regression, support vector machines, and random forests. Importantly, the diagnostic accuracy of the best performing model (RST) increased from 80.0% to 93.4% and diagnostic specificity increased from 57.2% to 93.4% when the MoCA score was combined with temporal metrics such as IRT-instrumental reaction time and TTS-submission time. This highlights that online platforms are able to detect subtle signs of bradykinesia (a hallmark symptom of Parkinson's disease) and use this as a biomarker to provide more precise and specific diagnosis. Despite the constrained number of participants (15 Parkinson's disease patients and 16 healthy controls), the results suggest that incorporating time-based metrics into cognitive screening algorithms may significantly improve their diagnostic capabilities. Therefore, these findings recommend the inclusion of temporal dynamics in MoCA assessments, which may potentially improve the early detection of NDs.