Deep Learning-Assisted Gait Parameter Assessment for Neurodegenerative Diseases: Model Development and Validation.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Clinical Diagnostics

Background:

• Neurodegenerative diseases (NDDs) pose a global health challenge, particularly for older adults.
• Early diagnosis of NDDs is difficult but critical for effective management and treatment.
• Gait analysis is a promising indicator for early NDD detection, offering insights for diagnosis, treatment, and rehabilitation.

Purpose of the Study:

• To develop a non-invasive, contactless gait assessment method using advanced machine learning.
• To provide healthcare professionals with precise gait parameters for improved NDD diagnosis and rehabilitation planning.
• To leverage artificial intelligence for a novel approach to gait evaluation.

Main Methods:

• Collected motion data from 41 participants (aged 25-85) using a 3D camera (Azure Kinect).
• Employed Support Vector Machine (SVM) and Bidirectional Long Short-Term Memory (Bi-LSTM) classifiers trained on spatiotemporal features.
• Utilized a 10-fold cross-validation strategy for model generalization and compared performance against a heuristic method.

Main Results:

• Bi-LSTM achieved superior classification performance with 90.54% precision, 90.41% recall, and 90.38% F1-score.
• Bi-LSTM demonstrated 93.2% accuracy in gait segmentation, significantly outperforming SVM (77.5%).
• The Bi-LSTM approach yielded a low average error rate of 3.17% in gait parameter calculation, compared to 5.85% for SVM and 20.91% for the heuristic method.

Conclusions:

• The Bi-LSTM-based contactless gait analysis effectively supports accurate gait parameter assessment.
• This AI-driven approach aids medical professionals in early NDD diagnosis.
• The method facilitates the creation of personalized and effective rehabilitation plans for NDD patients.