Digital Gait Biomarkers for Parkinson's Disease: Subject-Wise Validated Explainable AI Framework Using Vertical...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Neurology

Background:

• Parkinson's disease (PD) causes progressive gait issues, but current clinical scales are subjective and discrete.
• Accurate, continuous assessment of PD severity and early detection remain challenges.

Purpose of the Study:

• To develop and validate an explainable AI framework for reproducible PD detection and continuous severity estimation using gait data.
• To assess the potential of gait-based digital biomarkers for early PD screening and monitoring.

Main Methods:

• A two-stage AI framework was developed using vertical ground reaction force (VGRF) signals.
• Stage 1: Deep learning models (TCN, BiGRU, FCNN-Transformer) for PD detection.
• Stage 2: XGBoost regression for continuous severity estimation, validated with explainable AI (Integrated Gradients) and longitudinal data (TREND cohort).

Main Results:

• Deep learning models achieved high PD discrimination (AUC ≥ 0.93), with FCNN-Transformer outperforming others.
• Explainable AI identified gait variability as key PD indicator.
• AI accurately predicted continuous PD severity (Spearman ρ = 0.921, R² = 0.953).
• Gait changes were detected years before clinical diagnosis in a longitudinal cohort.

Conclusions:

• Gait-based digital biomarkers can objectively detect PD and quantify disease progression.
• The AI framework offers a reproducible, explainable, and clinically interpretable tool for PD assessment.
• This approach supports early screening and continuous monitoring of Parkinson's disease.