Machine learning-based gait adaptation dysfunction identification using CMill-based gait data.

Area of Science:

• Biomedical Engineering
• Rehabilitation Science
• Machine Learning in Healthcare

Background:

• Machine learning (ML) combined with gait analysis offers significant potential for diagnosing abnormal gait patterns.
• Gait adaptability is crucial for functional mobility and is often impaired following neurological conditions like stroke.

Purpose of the Study:

• To analyze gait adaptability characteristics in stroke patients.
• To develop and optimize ML models for identifying individuals with gait adaptability deficits (GAD).
• To identify key features crucial for classifying GAD in stroke patients.

Main Methods:

• Gait adaptability was assessed in 30 stroke patients and 50 healthy adults using a CMill treadmill during tasks like obstacle avoidance and speed adaptation.
• Multiple ML models, including AdaCost, were trained using demographic, kinematic, and adaptability data.
• Model performance was evaluated using accuracy, sensitivity, F1-score, and ROC-AUC.

Main Results:

• Stroke patients exhibited significantly reduced gait speed and step length, with increased asymmetry compared to healthy individuals.
• Gait adaptability tasks, particularly obstacle avoidance and speed adaptation, were significantly impaired in stroke patients.
• The AdaCost model achieved high classification performance (ACC: 0.85, SEN: 0.80, F1: 0.77, AUC: 0.75), identifying obstacle avoidance and gait speed as critical diagnostic features.

Conclusions:

• Stroke patients demonstrate impaired gait speed and adaptability, especially in obstacle negotiation and speed changes.
• Faster gait speed and better obstacle avoidance correlate with improved functional mobility in stroke survivors.
• ML models, particularly AdaCost, can effectively identify GAD, supporting clinical decision-making and the development of computer-aided diagnosis systems.