Identification of sensory information in mixed nerves using multi-channel cuff electrodes for closed loop neural prostheses

The addition of sensory feedback is expected to greatly enhance the performance of motor neuroprostheses. In the case of stroke or spinal cord injured patients, sensory information can be obtained from electroneurographic (ENG) signals recorded from intact nerves in the non-functioning limb. Here, we aimed to identify sensory information recorded from mixed nerves using a multi-channel cuff electrode. ENG afferent signals were recorded in response to mechanical stimulation of the foot corresponding to three different functional types of sensory stimuli, namely: nociception, proprioception and touch. Offline digital signal processing was used to extract features for use as inputs for classification. A quadratic support vector machine was used to classify the data and the five fold cross validation error was measured. The results show that classification of nociceptive and proprioceptive stimuli is feasible, with cross validation errors of less than 10%. However, further work is needed to determine whether the touch information can be extracted more reliably from these recordings.