Machine Learning Models for Parkinson Disease: Systematic Review.

Area of Science:

• Neurology
• Computer Science
• Biomedical Informatics

Background:

• Machine learning (ML) is increasingly applied to Parkinson disease (PD) detection and prediction due to data availability and advanced tools.
• Despite numerous studies, few ML systems for PD are adopted in clinical practice, often due to a lack of external validity.
• Methodological issues in ML design and reporting further impede the clinical integration of data-driven PD solutions.

Purpose of the Study:

• To systematically review current machine learning (ML) practices in Parkinson disease (PD) research.
• To assess the methodologies and reporting standards of ML applications for PD diagnosis and progression tracking.

Main Methods:

• A systematic literature review was conducted following PRISMA guidelines.
• PubMed database was searched for studies published between January 2020 and April 2021 using specific keywords related to Parkinson's and ML.
• 113 publications utilizing ML for PD classification or prediction were included in the analysis.

Main Results:

• Only 65.5% of studies employed a holdout test set to validate ML model accuracy.
• A significant number of studies (38.9%) failed to report model tuning procedures, with 27.4% using suboptimal ad hoc tuning.
• Direct model comparisons were reported in only 15% of studies, limiting result interpretability.

Conclusions:

• Current ML research for Parkinson disease exhibits notable methodological limitations.
• Inconsistent validation and reporting practices contribute to a performance gap between research findings and real-world clinical applicability.
• Addressing these limitations is crucial for the successful adoption of ML in PD detection and prediction.