Classification of Parkinson's Disease Gait Using Spatial-Temporal Gait Features

Area of Science:

• Biomedical Engineering
• Neurology
• Data Science

Background:

• Quantitative gait analysis is crucial for Parkinson's disease (PD) diagnosis and management.
• Patient physical properties and walking speed can obscure pathological gait features.

Purpose of the Study:

• To identify spatial-temporal gait differences between PD patients and controls using a multiple regression normalization strategy.
• To evaluate machine learning effectiveness in classifying PD gait after normalization.

Main Methods:

• Collected spatial-temporal gait data from 23 PD patients and 26 controls.
• Normalized data using dimensionless equations and multiple regression.
• Applied machine learning (Random Forest, SVM, KFD) for PD gait classification.

Main Results:

• Multiple regression normalization revealed significant differences in stride length, cadence, stance time, and double support time.
• Random Forest achieved 92.6% PD classification accuracy after multiple regression normalization.
• This surpassed accuracies from raw data or dimensionless equation normalization.

Conclusions:

• Multiple regression normalization effectively distinguishes PD gait features.
• This enhanced approach can aid in PD diagnosis and treatment using gait data.