Artificial neural networks based myoelectric control system for automatic assistance in hand rehabilitation

Myoelectric control is using electromyography (EMG) signal as a source of control, with this technique, we can control any computer based system such as robots, devices or even virtual objects. The tendon gliding exercise is one of the most common hand's rehabilitation exercises. In this paper, we present a patterns recognition based myoelectric control system (MCS) for the automatic assistance in tendon gliding exercise. The user is assisted by visual indicators and demo videos. EMG patterns recognition is done with EMG features and a multi-layer Artificial neural network (ANN), the ANN classifier output is used to synchronize the demo video with the detected movement, the transition between states is done automatically when the current state's movement is correct and the required number of repetition is reached. The ANN learning is done using back-propagation algorithm, we have used only two sEMG electrodes and four common used timedomain EMG feature extraction methods, the features quality is evaluated by the average Rand index using eight unsupervised clustering algorithms. The efficacy of the proposed method is experimentally validated with five able-bodied subjects, where we have reached an average classification accuracy of 95.11% and a processing time less than 300ms.