Cross-domain transfer learning strategy enhances interpretability of deep learning model explanations.
M, Z., A, H., H, M., & M, S. (2026). Cross-domain transfer learning strategy enhances interpretability of deep learning model explanations.. Scientific reports. https://doi.org/10.1038/s41598-026-59076-8
M Z, A H, H M, M S. Cross-domain transfer learning strategy enhances interpretability of deep learning model explanations.. Scientific reports. 2026; doi: 10.1038/s41598-026-59076-8
M Z, A H, H M, et al. Cross-domain transfer learning strategy enhances interpretability of deep learning model explanations.[J]. Scientific reports. 2026. DOI: 10.1038/s41598-026-59076-8.
@article{m2026,
author = {Zannini M and Hammer A and Malberg H and Schmidt M},
title = {Cross-domain transfer learning strategy enhances interpretability of deep learning model explanations.},
journal = {Scientific reports},
year = {2026},
doi = {10.1038/s41598-026-59076-8},
note = {PMID: 42342760},
}
TY - JOUR AU - Zannini M AU - Hammer A AU - Malberg H AU - Schmidt M TI - Cross-domain transfer learning strategy enhances interpretability of deep learning model explanations. T2 - Scientific reports PY - 2026 DO - 10.1038/s41598-026-59076-8 AN - PMID:42342760 ER -
Clinical decision-making increasingly relies on deep neural networks (DNNs), yet their deployment in practice requires transparent and interpretable predictions. Explainable artificial intelligence (xAI) methods can identify input regions relevant to a model's decision, but their clinical interpretability remains limited. In this study, we investigated whether inductive transfer learning (TL) can reinforce domain-specific feature separation in xECGArch, a two-branch convolutional neural network for atrial fibrillation (AF) detection from electrocardiograms (ECGs). Each branch was pre-trained on a task aligned with its designated feature domain, P wave detection for morphology and RR interval variability prediction for rhythm, then fine-tuned on binary AF classification using an iterative layer freezing schedule. Deep Taylor decomposition (DTD) was applied to analyze explanations across all configurations. Fine-tuning accuracy ranged from 85.70% to 95.23%, remaining comparable to the original xECGArch architecture and previous TL-based approaches. However, DTD analysis demonstrated that morphology pre-training directed relevance toward P waves, whereas rhythm pre-training concentrated explanations on R peaks, with domain specificity increasing as more layers were frozen. These findings suggest that inductive TL can encourage domain-specific feature attribution in DNNs, improving the alignment of post-hoc explanations with clinically meaningful ECG regions.