Background and Objective:​ Hand function assessments in a clinical setting are critical for upper limb rehabilitation after spinal cord injury (SCI) but are unable to accurately reflect performance in an individual’s home environment. When paired with computer vision models, egocentric videos from wearable cameras provide an opportunity for remote hand function assessment. This study presents a novel computer vision model for predicting clinical hand function assessment scores from egocentric video.

Methods:​ SlowFast, MViT, and MaskFeat models were trained and validated on a custom SCI dataset, which was annotated with clinical hand function assessment scores.

Results:​ An accuracy of 0.551±0.139 and F1 score of 0.547±0.151 was achieved on the 5-class classification task. An accuracy of 0.724±0.135 and F1 score of 0.733±0.144 was achieved on a consolidated 3-class classification task.

Conclusion:​ This novel algorithm, for the first time, demonstrates that automatically evaluating the quality of hand function from egocentric video is feasible.