Stroke Motor Imagery Recognition based on EEG

Motor imagery based brain computer-interface systems have been widely used in motor rehabilitation after stroke. An effective algorithm is very important whether it is for the analysis of the neuroplasticity results of brain injury or for the external mechanical skeleton to assist the movement of the affected side in patients with dyskinesia. This paper proposes a long short - term memory(LSTM)-Fully connected(FC)-LSTM model to enhance the learning ability of LSTM units. According to the characteristics of high time resolution of electroencephalography(EEG) signals, the lifting wavelet transform is introduced to remove noise, and the proposed framework is used to process the time series EEG signal. After adjusting the parameters of the model, we analyzed the EEG signals of 12 stroke patients while performing right/left hand motor imagination task, in order to improve the algorithm performance and to reduce the offline calibration time. The proposed framework aims to further localize brain injury area and to quantify brain injury degree.