MBBi-TCNet: Multi-branch bi-directional temporal convolutional network for EEG classification of mental imagery

• Published in: Biomedical signal processing and control Vol. 111; p. 108381
• Main Authors: Zhang, Zhun, Wang, Li, Li, Jin, Huang, Mingyang, Feng, Yujie
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.01.2026
• Subjects: Brain-computer interface (BCI); Deep learning; Electroencephalography (EEG); Mental imagery; Temporal convolutional network; Deep learning; Mental imagery; Brain-computer interface (BCI); Electroencephalography (EEG); Temporal convolutional network
• ISSN: 1746-8094
• DOI: 10.1016/j.bspc.2025.108381

The brain-computer interface (BCI) is a cutting-edge technology that makes the communication between the brain and external devices possible. However, due to the low signal-to-noise ratio and non-stationarity of electroencephalography (EEG) signals, the EEG-based BCIs encounter considerable challenges. In this study, a multi-branch bi-directional temporal convolutional network (MBBi-TCNet) is proposed for mental imagery decoding. By combining an attention mechanism module and a Bi-TCN module, the proposed model adopts a multi-branch network architecture to extract more useful features. By integrating the information bidirectionally, the long-range dependencies in sequences can be captured by the Bi-TCN module. The performance of the proposed model is assessed by utilizing the cross-validation method on three datasets. In the subject-dependent scenario, MBBi-TCNet achieves average classification results of 86.15 % (within-session) and 83.22 % (cross-session) on the BCIC IV2a dataset, and 80.97 % (within-session) and 86.53 % (cross-session) on the BCIC IV2b dataset, respectively. Additionally, it demonstrates an accuracy of 75.69 % (within-session) on the private dataset. In the subject-independent scenario, MBBi-TCNet also outperforms all baseline models. According to the classification results, MBBi-TCNet is superior to other state-of-the-art models. Therefore, the practical application of BCIs can be enhanced by the proposed model.