Individuals with spinal cord injuries experience reduced hand function, which diminishes theirindependence and quality of life. The current assessments conducted in clinical settings may not accurately reflect hand function in home and community environments. Evaluating hand function within a naturalistic environment can yield valuable information, which may result in more precise assessments of hand function. This research employs egocentric video to analyze hand function within a naturalistic context.

The literature on wearable technologies to measure impaired hand function in context has predominantly focused on the quantity of hand use, in contrast to describing the quality of hand movement. This thesis aims to bring forth innovative perspectives on hand function by quantifying the reliance on various grasp types among individuals with spinal cord injuries. Our research encompassed four key studies:​

1. The initial study highlighted the significance of grasp analysis during real activities of daily living at home which was found to be highly predictive of clinic-based assessments.
2. In the second study, we introduced an innovative computer vision and deep learning model to identify individuals relying heavily on a compensatory grasp called the tenodesis grasp.
3. The third study introduced a novel computer vision algorithm classifying hand grasps into power, precision, intermediate, and non-prehensile categories.
4. Finally, in the fourth study, we employed a semantic clustering approach to summarize hand grasps without restriction to predefined grasp types.

The studies in this thesis demonstrate for the first time automated analyses of impaired grasping after SCI outside of clinical settings. We believe that the findings from these studies can offer valuable insights into optimizing rehabilitation treatments to maximize hand function recovery.