Akademik Veri Yönetim Sistemi
Biomedical Signal Processing and Control, cilt.119, 2026 (SCI-Expanded, Scopus)
Swallowing is a vital motor function that ensures safe and efficient nutrient intake, yet it remains largely underrepresented in brain–computer interface (BCI) research, particularly in the domain of motor imagery (MI). In this study, we propose a novel deep learning framework that combines Convolutional Neural Networks (CNNs) with a transformer-based architecture to classify EEG signals corresponding to swallowing MI and resting states. The model leverages CNNs to extract local temporal features and applies self-attention mechanisms to capture long-range dependencies across EEG time series, enhancing both performance and interpretability. EEG data were collected from 30 healthy participants (aged 18–56) using a 16-channel wearable EEG cap at a 500 Hz sampling rate. Participants alternated between swallowing MI tasks and rest conditions. The classification performance of the proposed model was evaluated using a Leave-One-Subject-Out (LOSO) cross-validation approach to ensure subject independence. Our results demonstrate that the transformer-based model achieved the highest average classification accuracy (80.11% ± 9.88), precision (81.11% ± 10.65), and recall (80.67% ± 16.82), outperforming baseline models such as EEGNet, ShallowConvNet, and DeepConvNet. Statistical comparisons using the Friedman test with Bonferroni correction confirmed the model's significant superiority (p < 0.005). Additionally, t-SNE visualizations revealed distinct clustering between swallowing MI and rest conditions, suggesting robust discriminative feature learning. These findings underscore the potential of attention-based architectures in decoding complex oropharyngeal motor imagery from EEG. The proposed framework offers a promising foundation for future BCI-driven neurorehabilitation applications targeting swallowing disorders such as dysphagia.