A neural network approach to monitor motor activities

This study compares the performance of surface electromyography (EMG) with accelerometry (ACC) in the detection of functional motor activities. Its outcome will guide the development of a wearable system to automatically monitor an individual´s functional status in the home and relay that information to a remotely located caregiver. EMC and ACC signals were simultaneously recorded while subjects (N=11) performed a predetermined sequence of tasks. Two sets of tasks were considered: identification tasks related to clinical assessment of functional independence, and non-identification tasks namely motor activities biomechanically similar to the identification tasks but not related to clinical assessment. Pattern recognition algorithms were applied to data from both sensor types. The performance criteria were sensitivity (detection of identification tasks when present), specificity (avoidance of confusing identification tasks with one another), and misclassification (avoidance of confusing non-identification tasks with identification tasks). Results indicated that mean sensitivity over all tasks was higher for accelerometers than for EMG for a given value of misclassification. However, for specific tasks, EMG sensitivity was higher than that of accelerometers. Therefore we explored whether a combination of the two sensor types would lead to a further increase in sensitivity.