Abstract
In recent years, motor imagery electroencephalography decoding has become a promising research field in brain-computer interface. Motor imagery signals generated from the brain can be decoded into certain commands to control external devices. The application of some deep learning algorithms to motor imagery has shown good performance by increasing accuracy and stability, as deep learning can easily handle high-dimensional, non-linear, and non-stationary electroencephalogram data. In this paper, we reviewed trends and approaches of deep learning algorithms for motor based on previously published papers indexed in Web of Science. We screened thirty-six research papers of the motor imagery classification using deep learning methods in the period between 2010 and 2020. In addition, we found the closest terms of the study using the network visualization technique and word cloud. This paper summarizes different input formulation methods used for developing deep learning models. Finally, some suggestions and recommendations for future research on motor imagery classification have been proposed.