Spatial information considered convolutional neural network for electroencephalogram-based motor imagery classification

As brain-computer interface (BCI) technology continues to advance in various fields, it has become one of the possible solutions for patients with motor dysfunction who have healthy thinking ability to regain motor ability. The vigorous development of deep learning (DL) provides it with a possible tool to analyze electroencephalogram (EEG) signals. Through analyzing and categorizing EEG signals associated with motor imagery (MI), the system can effectively perceive the patient's motor intentions. Currently, Convolutional Neural Networks (CNN) have exhibited exceptional performance in a variety of fields, including computer vision (CV) and natural language processing (NLP). However, the brain structure has rich spatial information, which was not fully utilized by CNN for MI-EEG signal analysis in the past. This paper introduces SP-CNN, a convolutional neural network that incorporates spatial information from the brain, to address the classification challenge of MI-EEG signals. The experimental findings indicate that this method exhibits stable and robust performance across diverse subjects.