Temporal-Focal Attention on EEG for Cross-Patient Epileptic Seizure Detection

Epileptic seizures are sudden, brief episodes of abnormal brain activity that can be detected through electroencephalography (EEG), which records both normal and abnormal electrical activity in the brain. Previous studies have achieved significant success in detecting seizures using convolutional neural networks (CNNs) to automatically extract seizurespecific features from EEG signals. However, substantial variability in multi-channel seizure waveforms across different patients poses challenges to model generalization and cross-patient recognition. To this end, we introduce a novel Temporal-Focal Attention Network (TFANet), designed to capture channel-specific activations associated with crosspatient seizure patterns. TFANet enhances focal responses learned from high-resolution temporal signals during convolution, effectively capturing channel-specific seizure representations across patients. Extensive experiments on a self-curated dataset demonstrate that TFANet outperforms existing methods, achieving a state-of-the-art Kappa score of 93.19% for cross-patient seizure detection. The code is available at https://github.com/fjssharpsword/EEG-BCI.