Grasp Detection with Force Myography for Upper-extremity Stroke Rehabilitation Applications

Grasp training is a key aspect of stroke rehabilitation. This thesis explores the suitability of Force Myography (FMG) classification for the two-class problem of grasping, regardless of grasp-type, versus a lack of grasping, for rehabilitation applications. FMG-based grasp detection in individuals with stroke was assessed with a protocol comprising of three grasp-and-move tasks, requiring a single grasp-type. Accuracy was lower, and required more training data for individuals with stroke when compared to healthy volunteers. Despite this, accuracy was above 90% in individuals with stroke. FMG-based grasp detection was further evaluated using a second protocol comprising of multiple grasp-types and upper-extremity movements, with healthy volunteers. The utility of classifying temporal features of the FMG signal was also assessed using Area under the Receiver Operator Curve (AUC). Accuracy with the raw FMG signal was 88.8%. At certain window configurations, model-based temporal features yielded up to a 6.1% relative increase in AUC over the raw FMG signal.