Identifying Hand Use and Hand Roles After Stroke Using Egocentric Video

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Computer Vision

Background:

• Upper limb (UL) impairment is common after stroke, significantly impacting quality of life.
• Clinical assessments of UL function may not accurately reflect real-world activities of daily living (ADLs).
• Current home-based assessments rely on self-report and lack detailed hand function analysis, with accelerometers failing to capture fine hand movements.

Purpose of the Study:

• To develop and evaluate a wearable system using egocentric cameras and computer vision to identify hand use and hand roles during ADLs in stroke survivors.
• To differentiate between the manipulator and stabilizer roles of the more-affected hand in unconstrained environments.

Main Methods:

• Nine stroke survivors performed ADLs in a simulated home environment while wearing an egocentric camera.
• Features extracted included motion, hand shape, color, and size changes.
• Random forest classifiers were trained to detect hand use and classify hand roles, with evaluation using leave-one-subject-out and leave-one-task-out cross-validation.

Main Results:

• The system achieved F1-scores for more-affected hand use ranging from 0.64 to 0.76 and for less-affected hand use from 0.72 to 0.82.
• Hand role classification F1-scores were between 0.59 and 0.70, depending on the hand and cross-validation method.
• These results indicate robust prediction capabilities for hand use and roles.

Conclusions:

• Egocentric video analysis combined with computer vision is feasible for predicting hand use and hand roles in stroke survivors.
• This technology offers a promising approach for objective, detailed assessment of UL function in real-world settings.
• The findings support the development of advanced wearable systems for stroke rehabilitation and monitoring.