A Vision-Based Framework for Predicting Multiple Sclerosis and Parkinson's Disease Gait Dysfunctions-A Deep Learning...

Area of Science:

• Neurology and Biomedical Engineering
• Artificial Intelligence in Healthcare

Background:

• Gait dysfunction is a key symptom in neurological disorders like Multiple Sclerosis (MS) and Parkinson's Disease (PD).
• Accurate and accessible methods for gait analysis are crucial for early diagnosis and monitoring of these conditions.

Purpose of the Study:

• To evaluate a vision-based framework using multi-view digital cameras for predicting gait dysfunction in MS and PD patients.
• To compare the performance of various traditional machine learning and deep learning (DL) algorithms for gait classification.

Main Methods:

• Collected 3D gait data using multi-view digital cameras from individuals with MS, PD, and age-matched healthy older adults (HOA).
• Extracted 3D joint keypoints and applied a data-driven methodology for stride classification.
• Compared 16 different machine learning and DL algorithms, including residual neural networks and 1D convolutional neural networks (CNNs).

Main Results:

• Multi-scale residual neural network achieved perfect accuracy and AUC when generalizing from comfortable walking to walking-while-talking tasks.
• For subject generalization in comfortable walking, residual neural network showed the highest accuracy (78.1%) and AUC (0.87).
• 1D CNN demonstrated top performance in predicting gait dysfunction across different tasks and new subjects, achieving 79.3% accuracy and 0.93 AUC.

Conclusions:

• Deep learning models, particularly 1D CNNs and residual neural networks, show significant potential for predicting neurological gait dysfunction.
• This study highlights the viability of inexpensive, vision-based systems for diagnosing and monitoring conditions like MS and PD.
• The proposed framework offers a promising, non-invasive approach for neurological gait analysis.