Using EEG Data to Classify Hand Motor Imagery in Stroke Patients

EEG-driven brain-controlled prosthetics can greatly help people with movement disorders in their daily lives. Still, the task of accurately translating intentions into expected actions in the brain remains a technical challenge. In recent years, as a branch of machine learning, Deep Neural Networks (DNN) have made great progress in improving the recognition and classification ability of computer vision tasks. This study evaluate the information representation ability of 2 machine learning models and 16 NN models and their variants. The experiment is conducted on an open source EEG dataset of hemiplegic stroke patients, and we evaluate the thematic and cross-thematic performance of the above algorithm. We find that a single-layer GRU network remained an optimal choice in subject subject classification because it is able to effectively reduce model overfitting. For cross-subject analysis, some deep-learning models outperform all other methods due to the availability of training data and can function effectively in cross-subject decoding tasks. Independent Component Analysis (ICA) processing is also proved to be effective for improving the accuracy of classification. These improvements will provide important information for researchers in the field of Brain-computer interface (BCI) to select data processing strategies appropriate for their research tasks, it also lays a foundation for the future research on the interaction between the characteristics of bioelectric signals and the structure of neural networks.