Why the study?
Deployment of deep neural networks in clinical practice requires transparent predictions, but the clinical interpretability of explainable artificial intelligence methods remains limited.
Does inductive transfer learning improve domain-specific feature attribution and interpretability of deep neural networks for atrial fibrillation detection from ECGs?
Population
ECGs for AF detection
Comparison
Inductive transfer learning with domain-specific pre-training vs original xECGArch architecture and previous TL-based approaches
Design
Model development and validation study
Authors
Loading...
Transfer learning sustains xECGArch accuracy; leaves open prospective clinical validation before deployment.
Does inductive transfer learning improve domain-specific feature attribution and interpretability of deep neural networks for atrial fibrillation detection from ECGs?
Inductive transfer learning with domain-specific pre-training enhances the interpretability of deep learning models for AF detection by aligning post-hoc explanations with clinically meaningful ECG regions.
Zannini et al. (2026) studied this question.