Brain-computer interfaces (BCIs) enable direct communication between the brain and external devices, providing critical support for individuals with motor impairments. However, accurate motor imagery (MI) decoding from electroencephalography (EEG) remains challenging due to noise and cross-session variability. This study introduces EEG-MFTNet, a novel deep learning model based on the EEGNet architecture, enhanced with multi-scale temporal convolutions and a Transformer encoder stream. These components are designed to capture both short and long-range temporal dependencies in EEG signals. The model is evaluated on the SHU dataset using a subject-dependent cross-session setup, outperforming baseline models, including EEGNet and its recent derivatives. EEG-MFTNet achieves an average classification accuracy of 58.9% while maintaining low computational complexity and inference latency. The results highlight the model's potential for real-time BCI applications and underscore the importance of architectural innovations in improving MI decoding. This work contributes to the development of more robust and adaptive BCI systems, with implications for assistive technologies and neurorehabilitation.
Figures & tables
Fig. 1 : EEG-MFTNet’s architecture.The numbers displayed above the input and each component’s output (purple boxes) denote the corresponding output dimensions.
Model
Accuracy (%)
Parameters
Latency (ms)
EEGNet
53.7 ± 6.3
3,274
47.86
EEG-GENet
54.7 ± 8.5
7,146
48.06
AA-EEGNet
54.8 ± 8.9
517,498
48.83
EEG-MFTNet
58.9 ± 10.5
16,096
49.63
TABLE I : Comparison of classification accuracy (averaged across all sessions and subjects), model complexity, and inference latency (per trial) for various EEGNet-based architectures.
Fig. 2 : Subject-wise classification accuracy of EEG-MFTNet across the four testing sessions. Each cluster of bars represents a single subject, with each bar indicating the model’s accuracy during a specific testing session.
Transformer Encoder Stream
Multi-scale Conv Module
Accuracy (%)
–
–
53.7 *
–
✓
55.2
✓
–
57.2
✓
✓
58.9
TABLE II : Ablation study of EEG-MFTNet. Checkmarks (✓) indicate the inclusion of the corresponding component. “–” in both columns corresponds to the baseline EEGNet.
Fig. 3 : Topographic contribution maps (Gradient × Input) averaged over all correctly classified trials belonging to a specific class (left or right motor representation) for Subject 6, Session 4. Warm colors indicate greater channel importance.
Fig. 4 : EEG-MFTNet’s average prediction confidence during Class-specific electrode deletion tests. Blue: most important electrodes deleted; Orange: least important deleted; shaded areas show standard deviation.
Department of Computer Science City University of Hong Kong Hong Kong, China · Department of Computer Science Florida State University Tallahassee, United States