A Lightweight Machine Learning Framework for Post-Stroke Gait Abnormality Classification Using Wearable Gyroscope...

Area of Science:

• Biomechanics
• Machine Learning
• Rehabilitation Engineering

Background:

• Accurate gait abnormality classification is vital for stroke patient monitoring and rehabilitation.
• Wearable sensors offer a promising avenue for objective gait assessment.

Purpose of the Study:

• To develop a computationally efficient machine learning framework for distinguishing healthy from abnormal gait patterns.
• To evaluate the effectiveness of statistical gyroscope features for gait classification.

Main Methods:

• Extracted statistical z-axis angular velocity features from wearable gyroscope data of both limbs.
• Employed logistic regression, support vector machines, and ensemble methods for classification.
• Utilized a leave-one-out cross-validation strategy for robust performance evaluation.

Main Results:

• Several classifiers achieved accuracy and Area Under the Curve (AUC) values above 0.95.
• Random forest and support vector machine models demonstrated near-perfect class separability (AUC = 0.98).
• Minimal biomechanically relevant gyroscope features proved effective for gait classification.

Conclusions:

• The proposed lightweight machine learning framework accurately classifies gait abnormalities using gyroscope data.
• The system's computational efficiency makes it suitable for wearable and remote monitoring in healthcare.
• This approach enhances gait monitoring and rehabilitation strategies for stroke survivors.