An electromyographic (EMG) measuring system has been developed for controlling active upper limb orthoses for patients with Erbʼs palsy. Patients with Erbʼs palsy cannot move either their shoulders or elbows, but can move their wrists. Therefore, if they can move their shoulders and elbows by an orthosis, their activities of daily living (ADL) can be improved dramatically. To control the orthosis, EMGmeasurements of the four wrist movements of extension, flexion, ulnar deviation, and radial deviation are used. However, there is a risk of an erroneous motion, when a movement other than that intended is performed. Therefore, the discriminant error rate of the EMGmeasurement system should be reduced. Various channel configurations (eight, six, and four channels) were compared for measuring the EMG, and various machine learning techniques; namely, Euclidean distance (euc), Mahalanobis distance (mh), and a support vector machine (svm), were used to discriminate the movement. By defining a threshold voltage and degree of similarity, and by using the most preferred discrimination process, discriminant error rates decreased. Moreover, when the number of motions in training data for discrimination was increased from four to six by adding finger extension and flexion, the discriminant error rates using euc and svm also decreased. However, when the number of channels was reduced from eight to six or four, discriminant error rates increased. Overall, the svm method achieved the highest discriminant rate and lowest discriminant error rate. Finally, an online discrimination method was developed which achieved no difference in discriminant rate and output the results every 710 ms.
|ジャーナル||Transactions of Japanese Society for Medical and Biological Engineering|
|出版ステータス||Published - 2016|
ASJC Scopus subject areas