Epileptic Seizure Detection via Dynamic Graph Learning in EEG Signals EEG Sinyallerinde Dinamik Graf Ö?grenimi ile Epileptik Nöbet Tespiti
34th Signal Processing and Communications Applications Conference, SIU 2026, İstanbul, Türkiye, 7 - 10 Temmuz 2026, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/siu71813.2026.11636405
- Basıldığı Şehir: İstanbul
- Basıldığı Ülke: Türkiye
- Anahtar Kelimeler: dynamic connectivity, EEG, epilepsy, graph neural networks, uncertainty estimation
- Yıldız Teknik Üniversitesi Adresli: Evet
Özet
Epilepsy is a chronic neurological disorder affecting approximately 50 million people worldwide. This study proposes an end-to-end Dynamic Unrolled TVGL (DUTVGL) architecture capable of simultaneously modeling spatial and temporal dependencies in EEG signals. The architecture comprises 1D-CNN feature extraction, Variational Graph Structure Learning (V-GSL) for dynamic connectivity estimation, and GCN classification. Class imbalance is addressed through Focal Loss, while Evidential Deep Learning (EDL) provides prediction-level uncertainty estimation and a patient-specific self-calibration mechanism controls false alarm rates at the individual level. The model was evaluated on both CHB-MIT and Siena Scalp EEG datasets, achieving 96.4% accuracy with 98.0% sensitivity on CHB-MIT and 91.2% accuracy with 85.4% sensitivity on Siena via LOPO cross-validation. With only 116,040 parameters and an inference time of 24 ms, the model demonstrates suitability for real-time applications.