An Enhanced CTNet for Motor Imagery EEG Classification
DOI:
https://doi.org/10.54097/8benha05Keywords:
Brain-computer interface, Motor imagery, EEG, CTNet, Transformer, Stability-oriented training, Model fusionAbstract
To address the limited feature discriminability, substantial training fluctuations, and limited decision stability during inference in four-class motor imagery electroencephalogram (MI-EEG) classification, this paper proposes an enhanced CTNet method that integrates lightweight structural enhancement, stability-oriented training, and complementary fusion inference. Built upon CTNet, the proposed method introduces a lightweight token-channel gate to adaptively recalibrate high-level token features on top of the joint modeling of a convolutional front-end and a Transformer encoder, and further adopts a two-layer classification head to enhance final discriminative capability. During training, label smoothing, center loss, exponential moving average of parameters, Top-5 checkpoint averaging, composite-score-based model selection, and lightweight input augmentation are incorporated to reduce model fluctuations under small-sample conditions. During inference, the logits of the CTNet-Repro baseline model and the enhanced model are fused in a weighted manner to exploit their decision complementarity. Under the subject-dependent protocol on the BCI Competition IV 2a dataset, the proposed method achieves average Accuracy, Macro-F1, and Balanced Accuracy of 80.63%, 80.17%, and 80.41%, respectively, across 9 subjects, representing improvements of 4.52, 4.69, and 4.49 percentage points over CTNet-Repro. The results demonstrate that the proposed method can effectively improve discriminative performance, model-selection stability, and test-stage robustness for MI-EEG classification while largely preserving the original CTNet backbone. Highlights: (1) An enhanced CTNet is proposed for four-class MI-EEG classification.(2)A lightweight token-channel gate improves discriminative high-level token features. (3) Stabilized training combines label smoothing, center loss, EMA, and Top-5 averaging. (4) Composite-score-based model selection improves robustness under small-sample settings. (5) Logits-level post-fusion further boosts Accuracy, Macro-F1, and Balanced Accuracy.
Downloads
References
[1] Song Y, Zheng Q, Liu B, et al. EEG Conformer: Convolutional Transformer for EEG Decoding and Visualization[J]. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2023, 31: 710-719. DOI: 10.1109/TNSRE.2022.3230250.
[2] Zhang D, Li H, Xie J, et al. MI-CAT: A Transformer-Based Domain Adaptation Network for Motor Imagery Classification[J]. Neural Networks, 2023, 165: 451-462. DOI: 10.1016/j.neunet.2023.06.005.
[3] Liu K, Yang T, Yu Z, et al. MSVTNet: Multi-Scale Vision Transformer Neural Network for EEG-Based Motor Imagery Decoding[J]. IEEE Journal of Biomedical and Health Informatics, 2024, 28(12): 7126-7137. DOI: 10.1109/JBHI.2024.3450753.
[4] Zhao W, Jiang X, Zhang B, et al. CTNet: A Convolutional Transformer Network for EEG-Based Motor Imagery Classification[J]. Scientific Reports, 2024, 14: 20237. DOI: 10.1038/s41598-024-71118-7.
[5] Keutayeva A, Fakhrutdinov N, Abibullaev B. Compact Convolutional Transformer for Subject-Independent Motor Imagery EEG-Based BCIs[J]. Scientific Reports, 2024, 14: 25775. DOI: 10.1038/s41598-024-73755-4.
[6] Zhang J, Li K, Yang B, et al. Local and Global Convolutional Transformer-Based Motor Imagery EEG Classification[J]. Frontiers in Neuroscience, 2023, 17: 1219988. DOI: 10.3389/fnins.2023.1219988.
[7] Liu M, Liu Y, Shi W, et al. EMPT: A Sparsity Transformer for EEG-Based Motor Imagery Recognition[J]. Frontiers in Neuroscience, 2024, 18: 1366294. DOI: 10.3389/fnins.2024.1366294.
[8] Shi X, Li B, Wang W, et al. EEG-VTTCNet: A Loss Joint Training Model Based on the Vision Transformer and the Temporal Convolution Network for EEG-Based Motor Imagery Classification[J]. Neuroscience, 2024, 556: 42-51. DOI: 10.1016/j.neuroscience.2024.07.051.
[9] Altaheri H, Karray F, Karimi A H. Temporal Convolutional Transformer for EEG Based Motor Imagery Decoding[J]. Scientific Reports, 2025, 15: 32959. DOI: 10.1038/s41598-025-16219-7.
[10] Zhao W, Jiang X, Zhang B, et al. Multi-Scale Convolutional Transformer Network for Motor Imagery Brain-Computer Interface[J]. Scientific Reports, 2025, 15: 96611. DOI: 10.1038/s41598-025-96611-5.
[11] Chaudhary P, et al. A Two-Stage Transformer Based Network for Motor Imagery Classification[J]. Medical Engineering & Physics, 2024, 128: 104154. DOI: 10.1016/j.medengphy.2024.104154.
[12] Deny P, Cheon S, Son H, et al. Hierarchical Transformer for Motor Imagery-Based Brain Computer Interface[J]. IEEE Journal of Biomedical and Health Informatics, 2023, 27(11): 5459-5470. DOI: 10.1109/JBHI.2023.3304646.
[13] Sharma N, Upadhyay A, Sharma M, et al. Deep Temporal Networks for EEG-Based Motor Imagery Recognition[J]. Scientific Reports, 2023, 13: 18813. DOI: 10.1038/s41598-023-41653-w.
[14] Luo J, Wang Y, Xia S, et al. A Shallow Mirror Transformer for Subject-Independent Motor Imagery BCI[J]. Computers in Biology and Medicine, 2023, 164: 107254. DOI: 10.1016/j.compbiomed.2023.107254.
[15] Sartipi S, Cetin M. Subject-Independent Deep Architecture for EEG-Based Motor Imagery Classification[J]. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2024, 32: 718-727. DOI: 10.1109/TNSRE.2024.3360194.
[16] Li H, Zhang D, Xie J, et al. MI-DABAN: A Dual-Attention-Based Adversarial Network for Motor Imagery Classification[J]. Computers in Biology and Medicine, 2023, 152: 106420. DOI: 10.1016/j.compbiomed.2022.106420.
[17] Zhang D, Li H, Xie J, et al. MI-DAGSC: A Domain Adaptation Approach Incorporating Comprehensive Information of MI-EEG Signals[J]. Neural Networks, 2023, 167: 183-198. DOI: 10.1016/j.neunet.2023.08.008.
[18] She Q, Chen T, Fang F, et al. Improved Domain Adaptation Network Based on Wasserstein Distance for Motor Imagery EEG Classification[J]. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2023, 31: 1137-1148. DOI: 10.1109/TNSRE.2023.3241846.
[19] Zhong X C, Wang Q, Liu D, et al. A Deep Domain Adaptation Framework with Correlation Alignment for EEG-Based Motor Imagery Classification[J]. Computers in Biology and Medicine, 2023, 163: 107235. DOI: 10.1016/j.compbiomed.2023.107235.
[20] Li J, Shi J, Yu P, et al. Feature-Aware Domain Invariant Representation Learning for EEG Motor Imagery Decoding[J]. Scientific Reports, 2025, 15: 10664. DOI: 10.1038/s41598-025-95178-5.
[21] Li J, She Q, Meng M, et al. Three-Stage Transfer Learning for Motor Imagery EEG Recognition[J]. Medical & Biological Engineering & Computing, 2024, 62(6): 1689-1701. DOI: 10.1007/s11517-024-03036-9.
[22] Miao M, Yang Z, Sheng Z, et al. Multi-Source Deep Domain Adaptation Ensemble Framework for Cross-Dataset Motor Imagery EEG Transfer Learning[J]. Physiological Measurement, 2024, 45(5). DOI: 10.1088/1361-6579/ad4e95.
[23] Wang X, Liesaputra V, Liu Z, et al. An In-Depth Survey on Deep Learning-Based Motor Imagery Electroencephalogram (EEG) Classification[J]. Artificial Intelligence in Medicine, 2024, 147: 102738. DOI: 10.1016/j.artmed.2023.102738.
[24] Hameed I, Khan D M, Ahmed S M, et al. Enhancing Motor Imagery EEG Signal Decoding Through Machine Learning: A Systematic Review of Recent Progress[J]. Computers in Biology and Medicine, 2025, 185: 109534. DOI: 10.1016/j.compbiomed.2024.109534.
[25] Vafaei E, Hosseini M. Transformers in EEG Analysis: A Review of Architectures and Applications in Motor Imagery, Seizure, and Emotion Classification[J]. Sensors, 2025, 25(5): 1293. DOI: 10.3390/s25051293.
[26] Pfeffer M A, Wong J K W, Ling S H. Trends and Limitations in Transformer-Based BCI Research[J]. Applied Sciences, 2025, 15(20): 11150. DOI: 10.3390/app152011150.
[27] Song Y, Zheng Q, Wang Q, et al. Global Adaptive Transformer for Cross-Subject Enhanced EEG Classification[J]. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2023, 31: 2767-2777. DOI: 10.1109/TNSRE.2023.3285309.
[28] Wu R, Jin J, Daly I, et al. Classification of Motor Imagery Based on Multi-Scale Feature Extraction and the Channel-Temporal Attention Module[J]. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2023, 31: 3075-3085. DOI: 10.1109/TNSRE.2023.3294815.
[29] Pham T D. Classification of Motor-Imagery Tasks Using a Large EEG Dataset by Fusing Classifiers Learning on Wavelet-Scattering Features[J]. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2023, 31: 1097-1107. DOI: 10.1109/TNSRE.2023.3241241.
[30] Gómez-Morales Ó W, Collazos-Huertas D F, Álvarez-Meza A M, et al. EEG Signal Prediction for Motor Imagery Classification in Brain-Computer Interfaces[J]. Sensors, 2025, 25(7): 2259. DOI: 10.3390/s25072259.
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Journal of Computing and Electronic Information Management
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
How to Cite
