TSeizNet: Triplet Loss Empowered Multi-Scale CNN for Superior EEG Seizure Detection

Automated seizure detection using electroencephalography (EEG) reduces the workload for neurologists and accelerates treatment planning for epilepsy. However, this approach faces challenges due to the imbalance between seizure and non-seizure events, as well as the complexity of EEG setups. To address these issues, we propose TSeizNet, a novel multi-scale convolutional network utilizing triplet loss for EEG-based epileptic seizure detection. We evaluate TSeizNet on the CHB-MIT dataset under both full-scalp and low-density EEG settings. Experimental results show that TSeizNet achieves higher accuracy and F1-score compared to baseline methods, while maintaining acceptable sensitivity and specificity in both settings. These findings confirm the superior performance of TSeizNet and highlight the ability of low-density EEG to capture seizure-related content comparably with full-scalp EEG. This study suggests potential for future research in channel-reduced EEG for home monitoring and wearable device implementation.