DukeSpace will be down for maintenance at 9:15 AM EDT on Tuesday, July 29. Expected downtime is 20 minutes or less.

Show simple item record

dc.contributor.author Santa-Cruz, M. C. en_US
dc.contributor.author Riso, R. R. en_US
dc.contributor.author Lange, B. en_US
dc.contributor.author Sepulveda, F. en_US
dc.date.accessioned 2011-10-04T16:09:25Z
dc.date.available 2011-10-04T16:09:25Z
dc.date.issued 1999 en_US
dc.identifier.citation From "MEC 99," Proceedings of the 1999 MyoElectric Controls/Powered Prosthetics Symposium Fredericton, New Brunswick, Canada: August, 1999. Copyright University of New Brunswick. en_US
dc.identifier.uri http://hdl.handle.net/10161/4919
dc.description.abstract Hand prosthesis function is augmented when the user canemploy lateral grasp as well as traditional palmer grasp. Our goal in this investigation was to enable the below-elbow (BE) prostheses user to switch between and use these grasp modes in a natural and reliable manner. We recorded the EMG from residual muscles (flexor dig; ext dig; flex pollicis longus, ext pollicis longus) involved in these grasp activities in an adult subject with below elbow (BE) amputation while she contracted her residual forearm muscles to mimic computer animations of different hand movements. To reduce crosstalk between the recordings from seperate muscles, and to enhance the stability of the recording interface over the 30-day duration of the experimental sessions, we used chronically implanted percutaneous coiled wire electrodes implanted for 30 days (12 one-day sessions). Artificial Neural Network (ANN) pattern recognition techniques were used to extract voluntary command signals from the EMG signals. The mean absolute value (MAV) of the EMG signals was selected as a feature for training multilayer perceptions. Initially, we trained ANNs having 5 hidden neurons using data from the 10' and 12 session individually (3 training sessions each). Three additional ANNs (sizes 4:7:4, 4:8:4, 4:9:4) were designed and trained (3 training sessions each) with combined data from experimental sessions 10 and 12. Subsequently, we separately tested the performance of these ANNs with data from the 9, 10 and 12 experimental sessions. While the results showed that data from different experimental days were substantially consistent, more reliable recognition of the grasp mode from any arbitrary test sample (i .e.. taken from test sessions 9,10 or 12), was achieved when we used an ANN that was trained with representative samples from more than a single experimental day (e.g. using 10th and 12th experimental days data for training). This produced mean rates of recognition (averaged over the results from the three ANN training sessions with network size 4:8:4) of 97 6% key grip closing, 83 3% keygrip opening, 85.7% precision grip closing, 96.4% precision grip opening, for the combined evaluation data from all test sets. We conclude that intuitive operator selection, between key grip and precision grip modalities, is feasible for cases of BE amputation using recorded myoelectric signals. en_US
dc.publisher Myoelectric Symposium en_US
dc.title Natural Control Of Key Grip And Precision Grip Movements For A Myoelectric Prostheses en_US

Files in this item

This item appears in the following Collection(s)

Show simple item record