EEG Motor Imagery Classification using Integrated Transformer-CNN for Assistive Technology Control

Neuromuscular disorders encompass pathological conditions affecting either the musculoskeletal system, the ner-vous system, or their interface. The majority of neuromuscular disorders lack complete curative treatments and thus necessi-tate efforts focused on improving rehabilitation and quality of life. Assistive technologies with brain-computer interface (BCI) control provide a promising solution to main patients' daily activities. Central to these technologies is the classification of electroencephalography (EEG) motor imagery signals, which is paramount for the operational success of BCIs. The endeavor to enhance the precision of these classifications has led to the exploration of various computational techniques, with convo-lutional neural networks (CNNs) being at the forefront due to their proficiency in handling spatial hierarchies in data. Despite the advancements, the quest for improved accuracy continues, especially in scenarios characterized by raw EEG data inputs, as opposed to the transformed data inputs (e.g., using Fast Fourier Transform (FFT)) commonly reported in existing literature. This study introduces an integrated CNN model augmented with a transformer encoder block, aiming to harness the inherent spatial-temporal features of EEG signals more effectively. Preliminary evaluations suggest an encouraging comparable performance with the studies reported in the field, thereby underscoring its potential utility in enhancing motor imazery classification for BCI applications.