Key result
Guideline-based attention guidance fails to improve ECG model robustness over random lead masking alone.
Why the study?
Real-world 12-lead ECGs often have missing or corrupted leads. Whether injecting clinical knowledge via guideline-based attention guidance adds robustness to a deep-learning classifier beyond random lead masking during training was unknown.
Does adding guideline-based attention guidance to random lead masking during training improve the robustness of a deep-learning 12-lead ECG classifier to missing leads compared with random lead masking alone?
Does adding guideline-based attention guidance to random lead masking during training improve the robustness of a deep-learning 12-lead ECG classifier to missing leads compared with random lead masking alone?
Random lead masking during training is sufficient to make deep-learning ECG classifiers robust to missing leads, whereas constraining model attention to match clinical guideline leads adds no robustness and degrades probability calibration.
Answers are grounded in this paper and cite their sources.
Guideline attention guidance adds no robustness benefit beyond masking alone; leaves open whether alternative constraints or architectures could enhance ECG model resilience to missing leads.
Background Real-world 12-lead electrocardiograms (ECGs) often have missing or corrupted leads. A deep-learning classifier can be prepared for this by random lead masking during training, or by injecting clinical knowledge, for example by pulling its attention towards the leads that guidelines use for each diagnosis. Whether this knowledge adds robustness beyond masking is unknown. Methods Pre-registered comparison of three arms sharing one network (per-lead 1D-ResNet encoder and per-diagnosis attention over the 12 leads): standard training (M0); random whole-lead masking (M1); masking plus a Kullback-Leibler penalty pulling attention towards a fixed guideline-derived key-lead map (M2, guideline-based attention guidance). Six diagnoses; five seeds per arm. Development on PTB-XL (21,799 ECGs); single sealed evaluation on PTB-XL fold 10 (2,198 ECGs) and on CODE-15% (233,204 exams, external validation). Primary endpoint: AUROC difference M2 − M1, averaged over first-degree AV block, right and left bundle branch block and atrial fibrillation and over 1 to 6 lost (zeroed) leads, with a patient-level paired BCa bootstrap 95% confidence interval. Results The guidance did not change robustness: ΔAUROC −0.0010 (95% CI −0.0033 to +0.0006), an interval entirely within the pre-specified ±0.02 margin; the result was unchanged with another bootstrap seed, other interval methods and leave-one-diagnosis-out analyses. Masking improved AUROC: M1 − M0 = +0.0134 (95% CI +0.0102 to +0.0172) when one key lead was lost, and the number of lost leads tolerated before AUROC fell by more than 0.02 rose from 7 (M0) to 10 (M1). M2 attention matched the map and M2 was non-inferior on intact ECGs, but its calibration slope was lower for all six diagnoses in both datasets. M2 lost slightly more when all key leads were lost together (−0.0067) and gained only when one to three leads remained (exploratory; AUROC 0.9408 vs 0.9210 with 11 leads lost). CODE-15% reproduced every pattern (primary −0.0007, 95% CI −0.0011 to −0.0003). Conclusions Random lead masking suffices to make an ECG classifier robust to missing leads; adding guideline knowledge aligns attention with clinical criteria but adds no robustness and worsens calibration. Guideline-aligned attention should not be taken as evidence of robustness.
No takes yet. Share an insight, caveat, or question.
Ali et al. (2026) studied 12-lead ECG interpretation for cardiac conduction disturbances and arrhythmias (first-degree AV block, complete RBBB, complete LBBB, atrial fibrillation, sinus bradycardia, sinus tachycardia) (n=255,003). Guideline-based attention guidance plus random whole-lead masking (M2) vs. Random whole-lead masking alone (M1) and standard training without masking (M0) sharing the same per-lead 1D-ResNet encoder was evaluated on AUROC difference (M2 - M1) averaged over first-degree AV block, right and left bundle branch block, and atrial fibrillation across 1 to 6 lost (zeroed) leads on PTB-XL fold 10 (MD -0.0010, 95% CI -0.0033 to +0.0006, p=0.291). Guideline-based attention guidance added to random lead masking did not improve ECG model robustness to 1-6 missing leads compared to masking alone.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: