Machine learning approach to gait analysis for Parkinson's disease detection and severity classification

Area of Science:

• Neurology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Parkinson's disease (PD) is a progressive neurological disorder.
• Current PD severity assessment using the Hoehn and Yahr scale can be inconsistent, time-consuming, and costly.
• There is a need for objective and efficient methods to monitor PD progression.

Purpose of the Study:

• To develop and evaluate a machine learning-based gait classification system for predicting Parkinson's disease severity.
• To compare the performance of various machine learning algorithms in classifying PD stages.
• To enhance clinical decision-making through explainable AI for PD severity prediction.

Main Methods:

• Utilized two open-access datasets (PhysioNet, Figshare) containing ground reaction force data from PD patients.
• Applied machine learning algorithms: Decision Tree, Random Forest, Extreme Gradient Boost, and Light Gradient Boosting Machine (LGBM).
• Employed Explainable Artificial Intelligence (XAI) to interpret LGBM classification pathways for PD severity (Hoehn and Yahr scale 0-5).

Main Results:

• The Light Gradient Boosting Machine (LGBM) demonstrated superior performance, achieving 98.25% accuracy, 98.35% precision, 98.25% recall, and 98% F1 score on dataset 1.
• Performance on dataset 2 showed slightly lower but still robust results: 85% accuracy, 95% precision, 85% recall, and 89% F1 score.
• XAI provided insights into the LGBM classifier's decision-making process for PD severity prediction.

Conclusions:

• The developed machine learning system effectively assists in classifying Parkinson's disease severity using gait analysis.
• LGBM shows significant potential for automated and rapid screening of PD patients.
• Future work could involve integrating wearable sensors for real-time monitoring systems.