Flexible Machine Learning Algorithms for Clinical Gait Assessment Tools.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Clinical Biomechanics

Background:

• Current gait diagnostics rely on expensive motion-capture labs and expert staff.
• Wearable sensors and machine learning can enhance accessibility but often need large, manually labeled datasets.
• Existing algorithms lack flexibility and require extensive training data.

Purpose of the Study:

• To validate a novel machine learning algorithm for automated gait partitioning.
• To assess the algorithm's performance on both laboratory and wearable sensor data.
• To demonstrate reduced training data requirements for gait analysis.

Main Methods:

• Developed an artificial intelligence tool combining reinforcement learning and deep neural networks.
• Tested the algorithm on patients with central neurological lesions and severe gait impairments.
• Required only 2-3% of the dataset for training, significantly less than traditional methods.

Main Results:

• Achieved mean errors of 0.021 s for lab data and 0.034 s for sensor data.
• Demonstrated that <5% of the dataset is sufficient for training the AI model.
• The AI tool enabled automated gait partitioning with high accuracy.

Conclusions:

• The novel AI algorithm significantly reduces the need for extensive training data (<5%) for gait analysis.
• This approach enhances the flexibility and accessibility of objective gait assessments in clinical settings.
• Non-experts can readily adapt the algorithm for various measurement systems and patient populations.