Compact convolutional transformer for subject-independent motor imagery EEG-based BCIs

• Published in: Scientific reports Vol. 14; no. 1; pp. 25775 - 11
• Main Authors: Keutayeva, Aigerim, Fakhrutdinov, Nail, Abibullaev, Berdakh
• Format: Journal Article
• Language: English
• Published: London Nature Publishing Group UK 28.10.2024; Nature Publishing Group; Nature Portfolio
• Subjects: 631/378/116/1925; 631/378/116/2394; 631/378/116/2396; 639/166/985; 639/705/117; Algorithms; Brain-Computer Interfaces; Brain–computer interface; Compact convolutional transformers; Deep learning; EEG; Electroencephalography; Electroencephalography - methods; Humanities and Social Sciences; Humans; Imagination - physiology; Mental task performance; Motor imagery; multidisciplinary; Neural signal processing; Science; Science (multidisciplinary); Compact convolutional transformers; Deep learning; Motor imagery; EEG; Brain–computer interface; Neural signal processing
• ISSN: 2045-2322; 2045-2322
• DOI: 10.1038/s41598-024-73755-4

Motor imagery electroencephalography (EEG) analysis is crucial for the development of effective brain-computer interfaces (BCIs), yet it presents considerable challenges due to the complexity of the data and inter-subject variability. This paper introduces EEGCCT, an application of compact convolutional transformers designed specifically to improve the analysis of motor imagery tasks in EEG. Unlike traditional approaches, EEGCCT model significantly enhances generalization from limited data, effectively addressing a common limitation in EEG datasets. We validate and test our models using the open-source BCI Competition IV datasets 2a and 2b, employing a Leave-One-Subject-Out (LOSO) strategy to ensure subject-independent performance. Our findings demonstrate that EEGCCT not only outperforms conventional models like EEGNet in standard evaluations but also achieves better performance compared to other advanced models such as Conformer, Hybrid s-CViT, and Hybrid t-CViT, while utilizing fewer parameters and achieving an accuracy of 70.12%. Additionally, the paper presents a comprehensive ablation study that includes targeted data augmentation, hyperparameter optimization, and architectural improvements.