Machine Learning-Based Classification of Dependence in Ambulation in Stroke Patients Using Smartphone Video Data

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Artificial Intelligence in Healthcare

Background:

• Stroke survivors often experience impaired mobility and balance, affecting their ambulation independence.
• Accurate assessment of ambulatory dependence is crucial for effective rehabilitation and patient safety.
• Current methods may be time-consuming or require specialized equipment, limiting real-time monitoring.

Purpose of the Study:

• To develop and validate a deep learning framework for classifying ambulatory dependence in stroke patients.
• To utilize smartphone-recorded video data for objective assessment of ambulation during inpatient rehabilitation.
• To integrate gait parameters like swing time asymmetry for enhanced classification accuracy.

Main Methods:

• A 3D convolutional neural network (3D-CNN) model was developed using video clips of stroke patients' ambulation.
• Functional Ambulatory Categories (FACs) and Berg Balance Scale (BBS) scores were used as ground truth for dependence.
• Patient-centered video and pose data were extracted; swing time asymmetry was calculated and combined with 3D-CNN analysis.

Main Results:

• The 3D-CNN model achieved 86.3% accuracy, 94.0% recall, and 90.5% F1 score when trained on FACs and BBS.
• Integrating swing time asymmetry improved performance to 88.7% accuracy, 95.7% recall, and 92.2% F1 score.
• The framework demonstrated robust classification of ambulatory dependence using readily available video data.

Conclusions:

• The proposed deep learning framework effectively classifies ambulatory dependence in stroke patients using smartphone videos.
• This technology can alert clinicians and caregivers to patients requiring assistance, enhancing safety during unsupervised ambulation.
• The framework supports personalized rehabilitation strategies by objectively monitoring mobility and balance function.