Computational Modalities in Parkinson’s Disease Diagnosis: Advances in Methods, Challenges in Application, and Future Research Directions

Abstract

Background: Parkinson's disease affects over 11 million people worldwide, with an age-standardized incidence of 15.63 new cases per 100,000. Current diagnostic approaches rely on clinical evaluation and expensive neuroimaging techniques, which are a major obstacle to early detection and timely intervention. Artificial intelligence technologies offer unprecedented opportunities for the development of objective and affordable diagnostic systems, but there are still significant gaps between current evidence and clinical application requirements.
Objective: The systematic review synthesizes the evidence on computational modalities in PD diagnosis approaches, assesses the diagnostic accuracy and clinical feasibility of different modalities and algorithms, and identifies critical obstacles to clinical translation.
Methods: The comprehensive literature review of 51 studies (2020-26) examining the use of computational modalities in the diagnosis of PD. The studies were categorized by diagnostic method including (neuroimaging, behavioral, biochemical), artificial intelligence methodology (deep learning, classical ML, explainability approaches), and quality metrics including data leakage, external validation and bioavailability.
Key Findings: Ten diagnostic modalities demonstrated varying performance: SPECT DaTSCAN achieved 98.47% accuracy but faces accessibility constraints, while emerging modalities including speech analysis (89-95% accuracy), gait assessment (86.7-87.65% accuracy), and multimodal EEG+MRI fusion (97.17% accuracy) offer greater accessibility. However, critical methodological issues were endemic: data leakage affected 87% of neuroimaging AI studies, only 8% employed external validation, and 48% used inappropriately complex models for dataset size. Gender-stratified performance gaps of 10-20% documented substantial sex bias. Explainability methods including SHAP, attention heatmaps, and cross-attention mechanisms enhance clinical interpretability while maintaining diagnostic accuracy.
Implications: Consequences Although AI has significant diagnostic potential in multiple modalities, its translation into clinical practice requires: (1) prospective external validation in independent populations, (2) standardized data acquisition and reporting protocols, (3) development of gender-specific models, (4) large-scale multi-center studies (more than 1,000 subjects), and (5) health economics. The current evidence is insufficient to recommend any artificial insemination system for routine clinical use without robust prospective validation.
Conclusion: Artificial intelligence has emerged as a promising diagnostic tool for multiple sclerosis with a variety of modalities and robust reported precision metrics. However, the systematic methodological constraints and the lack of prospective clinical validation are critical barriers to the uptake of the clinical trial. Future research must prioritize external validation, quality standardization, equity considerations, and real-world implementation studies to bridge the gap between research evidence and clinical practice.