A Vision-Based System for Stage Classification of Parkinsonian Gait Using Machine Learning and Synthetic Data

Area of Science:

• Biomedical Engineering
• Neurology
• Computer Vision

Background:

• Parkinson's disease (PD) significantly impacts gait, worsening with disease progression.
• Current pose-estimation and machine learning methods classify gait but struggle with analyzing its progression for disease staging.
• Limited gait data hinders the clinical application of gait analysis technology for Parkinson's disease.

Purpose of the Study:

• To develop a quantitative, vision-based prognosis method for Parkinsonian gait severity using limited data.
• To facilitate the study and rehabilitation of Parkinsonian gait.
• To compare the performance of k-nearest neighbors (KNN), support-vector machine (SVM), and gradient boosting (GB) algorithms in classifying gait features.

Main Methods:

• A vision-based system was developed to analyze Parkinsonian gait at various stages.
• Linear interpolation of Parkinsonian gait models was employed.
• Performance comparison of KNN, SVM, and GB algorithms for gait feature classification.

Main Results:

• The proposed system achieved high accuracy in evaluating Parkinsonian gait prognosis.
• Classification accuracy ranged from 96% to 99% across tested algorithms.
• The study demonstrated the effectiveness of the vision-based approach for gait severity assessment.

Conclusions:

• The developed vision-based system offers an accurate method for assessing Parkinsonian gait severity.
• This approach can aid in the quantitative prognosis and rehabilitation of Parkinson's disease.
• The system's high accuracy, even with limited data, supports its potential clinical utility.