Teaching an ECG model where to look does not make it more robust to missing leads: a pre-registered comparison of guideline-based attention guidance with random lead masking
- 1Faculty of Medicine of Monastir, University of Monastir, Monastir, Tunisia
Abstract
BackgroundReal-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.
MethodsPre-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.
ResultsThe guidance did not change robustness: {Delta}AUROC -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).
ConclusionsRandom 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.
Related
- HAFI-DTA: an interpretable multimodal framework for drug-target affinity prediction based on hierarchical attention guidance and feature iterative interaction
- Prospective Comparison of Cognitive MRI-Targeted Biopsy Under Micro-Ultrasound Guidance Versus MRI/Ultrasound Fusion-Guided Transperineal Biopsy for Clinically Significant Prostate Cancer Detection
- Individualized and Platform-Designated Gene-Editing Studies in the Registered Clinical Record: A Case-by-Case Census Against 148 Eligible Studies
- Global clinical development of transcranial temporal interference stimulation: an analysis of registered human trials
- The 2026 British Society for Rheumatology guideline for the treatment of polymyalgia rheumatica
BibTeX
@misc{bou2026teaching,
title = {{Teaching an ECG model where to look does not make it more robust to missing leads: a pre-registered comparison of guideline-based attention guidance with random lead masking}},
author = {M. N. BOU ALI and M. Mahmoud},
howpublished = {medRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.17.26365795},
url = {https://www.medrxiv.org/content/10.64898/2026.10.17.26365795v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - Teaching an ECG model where to look does not make it more robust to missing leads: a pre-registered comparison of guideline-based attention guidance with random lead masking
AU - M. N. BOU ALI
AU - M. Mahmoud
PY - 2026
DA - 2026/10/17
PB - medRxiv
DO - 10.64898/2026.10.17.26365795
UR - https://www.medrxiv.org/content/10.64898/2026.10.17.26365795v1
AB - BackgroundReal-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. MethodsPre-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. ResultsThe guidance did not change robustness: {Delta}AUROC -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). ConclusionsRandom 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.
ER -