Detecting upper extremity activity with force myography

A novel technique named force myography (FMG) is shown to be able to detect some upper extremity (UE) status. It holds the potential to be a cost-effective solution to detect complex UE movements, which can be used for UE status monitoring and human-machine interface applications. In this thesis, a novel FMG system to capture the forearm muscle movement information was proposed. The system’s capability to detect complex UE postures was investigated. Also, its feasibility to be used in real-time UE posture detection applications, and its applicability to be utilized in a non-real-time UE activity tracking scenario were examined. Specifically, the capability of using FMG to predict the elbow, forearm, and wrist positions was investigated by studying the corresponding FMG signal patterns and classification performances. This study also used the more established surface electromyography (sEMG) method to identify the strength and weakness of FMG. Elbow, forearm, and wrist position predictions using FMG achieved 84%, 82%, and 71% accuracies respectively. The sEMG method yielded 75%, 83% and 92% accuracies for predicting the same respective positions. The feasibility of using FMG to predict a set of complex UE postures in real-time was investigated using a custom designed classification system. An experiment which required volunteers to perform a sequence of UE postures simulating a functional action, i.e., drinking from a cup, was conducted to examine the classification performance. An average accuracy of 92% with standard deviation of 3% was obtained from 6 volunteers. Also, the same system was used for controlling a robotic device that assists the forearm rotation. A 96% accuracy for predicting five forearm positions for controlling the device was obtained. Finally, the applicability of using FMG to count grasping actions during a pick-and-place (PAP) exercise was investigated. Two wireless FMG straps were prototyped to enable the capturing of FMG signal during the arm movement. A median percentage error of 1% with an interquartile range of 5% was achieved for counting 120 PAP actions with ten volunteers.