Creative Commons source reading · Explainable decision support
Cross-domain transfer learning strategy enhances interpretability of deep learning model explanations
Zannini M; Hammer A; Malberg H; Schmidt M
Originally published in Sci Rep · 2026-06-24
From the article
Selected source text. JournalOS removes markup and shortens excerpts where indicated. The full article provides the complete context.
Abstract
AbstractClinical 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…
Discussion
The features of AF can be partitioned into two domains, a rhythmic and a morphological one. Cardiologists assess the presence of AF based on such distinct feature domains: the absence of P waves and the presence of fibrillatory waves represent morphological indicators, while the irregular timing of heartbeats constitutes the rhythmic hallmark. The architecture of xECGArch reflects this clinical reasoning by assigning the STM to morphological features and the LTM to rhythmic features. However, the learned representations may overlap across the two branches, potentially undermining the interpretability of model explanations. If the ECG segments highlighted by xAI methods cannot be clearly attributed to either the morphological or the rhythmic domain, clinicians cannot reliably trace the model’s reasoning back to established diagnostic…
Editorial reading guide
- What data created and tested the model, and was evaluation independent?
- What is the cost of a false positive, false negative or out-of-distribution decision?
