Data-Driven Identification of Stroke through Machine Learning Applied to Complexity Metrics in Multimodal...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Rehabilitation Science

Background:

• Stroke causes brain damage, often leading to gait impairments requiring rehabilitation.
• Electromyography (EMG) and stereophotogrammetry assess muscle activation and movement during walking.
• Complexity metrics of EMG and kinematics can identify pathological gait patterns.

Purpose of the Study:

• To differentiate individuals with stroke from healthy controls using machine learning.
• To analyze gait complexity using EMG and kinematics data.
• To evaluate the effectiveness of machine learning classifiers for stroke gait assessment.

Main Methods:

• Collected EMG and kinematics data during walking.
• Applied complexity metrics (sample entropy, approximate entropy, spectral entropy) to EMG and kinematics.
• Utilized machine learning classifiers, including cubic support vector machine, for classification.
• Calculated conditional entropy between EMG and kinematics to analyze relationships.

Main Results:

• Machine learning classifiers effectively distinguished between stroke patients and healthy controls.
• The cubic support vector machine applied to EMG complexity metrics achieved 99.85% accuracy.
• Complexity measures of gait, muscle activation, and their interrelationships provided significant discriminatory information.

Conclusions:

• Machine learning analysis of gait complexity metrics derived from EMG and kinematics is a highly accurate method for stroke detection.
• This approach shows promise for objective monitoring of motor recovery in stroke patients.
• The findings support the integration of advanced computational methods in clinical gait analysis for stroke rehabilitation.