Jan–Jun 2025 · Jan–Jun 2026
ECG-Based Patient-Independent Epileptic Seizure Detection via Beat-to-Beat Dynamic Distribution Modeling and Label Uncertainty Suppression
USTC · First author · Prepared for submission to Journal of Neural Engineering
- Developed a lightweight patient-independent ECG seizure-detection framework using beat-to-beat waveform and heart-rate dynamic-distribution features.
- Designed an EFCNN-based classifier with weighted cross-entropy to reduce the impact of uncertain seizure labels.
- Achieved 84.6% seizure-level true detection and a 0.138 h⁻¹ false-alarm rate under leave-one-subject-out evaluation, reducing false alarms by 94% over the reproduced state-of-the-art baseline.