Gait-based Parkinson's disease diagnosis and severity classification using force sensors and machine learning

Area of Science:

• Biomedical Engineering
• Computational Neuroscience
• Medical Informatics

Background:

• Parkinson's disease (PD) is a progressive neurodegenerative disorder impacting motor function.
• Early detection and monitoring of PD progression are challenging due to subtle mobility changes and limitations of conventional methods.
• Gait analysis using force sensors offers a promising avenue for objective assessment of motor deficits.

Purpose of the Study:

• To propose a novel dual-stage machine learning model for classifying Parkinson's disease (PD) and assessing its severity.
• To leverage gait signal analysis from force sensors for enhanced diagnostic accuracy.
• To improve early detection and monitoring of PD progression.

Main Methods:

• A dual-stage model employing a hypertuned Random Forest Tree (RFT) for PD classification and an Ensemble Regressor (ER) for severity prediction.
• Utilized Vertical Ground Reaction Force (VGRF) sensor data from 166 participants (93 PD, 73 controls).
• Applied data balancing (SMOTE), feature extraction (time, frequency, spatial, temporal domains), and Recursive Feature Elimination (RFE) for optimal feature selection.

Main Results:

• The RFT classifier achieved 97.5% accuracy, 97% sensitivity, and 95% specificity in differentiating PD from non-PD participants.
• The ER regressor predicted disease severity with 96.4% accuracy, 0.065 mean absolute error, and 0.080 root mean square error.
• The proposed model demonstrated superior performance compared to existing methodologies.

Conclusions:

• The dual-stage model is highly effective for the early detection and severity assessment of Parkinson's disease.
• Gait analysis combined with advanced machine learning provides a robust and accurate approach for PD management.
• The findings suggest significant potential for clinical application in improving patient outcomes.