Convolutional Transformer Network for Motor Imagery Finger Classification in EEG-Based BCI

Mohsen Eslami, Kamran Kazemi, Amirhossein Adibpour, Ardalan Aarabi, [e-mail]

Keywords: Motor Task Differentiation, Neural Signal Decoding, Movement Imagination

Abstract: Brain-computer interface (BCI), which decode neural signals such as electroencephalograph (EEG), provides a direct communication bridge between human brain and external devices. Motor imagery (MI) relies on the mental simulation of motor actions without actual physical execution. MI-BCI enable control of external devices through imagined movements and demonstrated significant potential in neurorehabilitation for stroke patients and robotics control. However, the reliable classification of fine motor tasks, such as differentiating individual finger movements within the same hand, remains highly challenging due to overlapping regions in the sensorimotor cortex. This overlap complicates their differentiation from noninvasive recordings and often results in low classification accuracy. To address this, we propose a convolutional neural network–based method to improve the discrimination between thumb and little finger movements.

Submission Number: 54