Hand gestures recognition using electromyography for bilateral upper limb rehabilitation

Among the rehabilitation techniques used for upper limb stroke rehabilitation, bilateral training assisted by functional electrical stimulation may provide improved functional benefit. In these protocols, contralateral movement is often measured using kinematic sensors, which may not fully estimate the volitional effort performed by the patient. In this work, we propose the use of an easy-to-use wearable electromyography system to detect hand gestures in such bilateral therapies. Classification is performed using a combination of techniques, such as dimension reduction, clustering and support vector machines, in order to possibly remove the need for retraining the classifier for new users. Preliminary evaluation on 5 healthy subjects indicated 81% classification performance for the 5 gestures classifier, while using the 9 gestures classifier 46% of gestures were correctly classified, which may indicate limitations of the proposed method to hand poses that present high biomechanical similarity.