Cross-Subject Prediction in ECGI using a MARS-Based Regression Framework
52nd International Computing in Cardiology, CinC 2025, Sao-Paulo, Brezilya, 14 - 17 Eylül 2025, cilt.52, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Cilt numarası: 52
- Doi Numarası: 10.22489/cinc.2025.367
- Basıldığı Şehir: Sao-Paulo
- Basıldığı Ülke: Brezilya
- Orta Doğu Teknik Üniversitesi Adresli: Evet
Özet
Noninvasive electrocardiographic imaging (ECGI) reconstructs epicardial electrical activity from body surface potentials (BSP). In previous work, we demonstrated that a regression framework based on Multivariate Adaptive Regression Splines (MARS) trained within a subject achieved higher fidelity than Tikhonov regularization. In this study, we extend this framework to assess cross-subject generalization. A leave-one-pig-out protocol was applied on in-vivo data from four anesthetized pigs with torso and epicardial recordings (184 BSP and 239 EGM electrodes). To enable cross-subject training, BSP and EGM were standardized to 2D electrode maps. Results showed that the generalized MARS model maintained accuracy close to subject-specific training, and in some cases outperformed it when training data within the test subject were limited. These findings highlight the potential of data-driven ECGI frameworks to move beyond subject-specific models and towards clinically transferable solutions.