Clinical-Aware CNN with 1D Grad-CAM for AFib Detection on Wearable ECG Giyilebilir EKG için Klinik Odakli CNN ve 1B Grad-CAM ile AFib 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/siu71813.2026.11636968
- Basıldığı Şehir: İstanbul
- Basıldığı Ülke: Türkiye
- Anahtar Kelimeler: atrial fibrillation, cost-sensitive learning, edge AI, Grad-CAM, wearable ECG
- Gazi Üniversitesi Adresli: Evet
Özet
We present a lightweight, edge-compatible convolutional neural network with approximately 66 thousand parameters for four-class atrial fibrillation detection on the PhysioNet/CinC 2017 single-lead ECG dataset. A cost-sensitive loss function penalizes false negatives via an asymmetric weight, meeting clinical safety targets. As a result of ten-fold stratified cross-validation, our model achieves a clinically competent atrial fibrillation sensitivity, substantially outperforming a wavelet scattering transform and random forest-based baseline model. 1D Grad-CAM analyses confirm that the model focuses on clinically meaningful markers such as irregular R-R intervals and absent P-waves.