Leveraging Cardiac Imaging to Improve ECG-Based Detection of Chagas Disease in Resource-Constrained Settings
Laura Alvarez-Florez, Daniel Uyterlinde, Samuel Ruipérez-Campillo, Lukas P. A. Arts, Folkert W. Asselbergs, Fleur V. Y. Tjong
cs.LG · cs.AI · eess.IV
Quick verdict
Claims CMR-aligned ECG pretraining lifts Chagas detection (AUROC 0.827→0.851, frozen probe); credible, but lacks significance testing and alignment baselines.
Summary
Drawing on 63,193 paired ECG-CMR examinations from the UK Biobank, this paper transfers structural cardiac knowledge from imaging to ECG through contrastive pre-training, aiming to improve Chagas disease detection where MRI is scarce. A frozen CineMA CMR encoder's end-diastolic and end-systolic embeddings are distilled through a supervised MLP that regresses seven phenotypes (LVEF, LVEDV, LV mass, RVEDV, LVGLS, AFib, MI) into a 256-dimensional target space; an ECG-FM layer-9 representation is then aligned to it with an asymmetric InfoNCE loss (tau=0.1). The aligned encoder is probed frozen for Chagas on CODE-15% plus SaMi-Trop (~345,000 ECGs, 8,192 positives, 2.4% prevalence): five-fold cross-validation gives AUROC 0.851 and Top5%-TPR 0.427 versus 0.827 and 0.377 for unaligned ECG-FM, and under the challenge protocol Top5%-TPR 0.431 versus 0.381 for the PhysioNet 2025 winner's linear probe. On hidden test sets the submission scores 0.269 overall, below the top three teams (0.280-0.323), but posts the best SaMi-Trop-3 AUROC (0.773) and ELSA-Brasil challenge score (0.132) among them. Main limitations: the headline AUROC gain lacks significance testing; prior ECG-CMR alignment methods (MMCL, PTACL, ECCL) are cited but not compared; the target-space design is not ablated; no code release is stated; and the best ELSA-Brasil challenge score coexists with the worst ELSA AUROC (0.556) among compared methods.
Review complete
Assessment at a glance
5
risk flags identified
● high 0
● medium 3
● low 2
5 strengths · 5 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 20 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- Contrastive alignment of an ECG encoder to a clinically grounded CMR embedding space improves ECG-based Chagas detection even though no Chagas cases or Chagas-related imaging appeared in pre-training.
- A frozen linear probe on the aligned encoder reaches AUROC 0.851 and Top5%-TPR 0.427 on CODE-15% plus SaMi-Trop in five-fold cross-validation, versus 0.827 and 0.377 for unaligned ECG-FM (Table 1).
- On the hidden PhysioNet/CinC 2025 test sets the submission attains the highest SaMi-Trop-3 AUROC (0.773) and best ELSA-Brasil challenge score (0.132) among the top three teams, despite ranking fourth overall (0.269 vs 0.280-0.323).
- CMR is required only during pre-training; deployment runs on a standard 12-lead ECG, preserving applicability in resource-constrained settings.
Strengths
Stringent cross-disease transfer test
Chagas patients and Chagas imaging are entirely absent from UK Biobank pre-training (Sections 1, 3.4); evaluating with a frozen linear probe isolates representation quality from fine-tuning capacity, and deployment is ECG-only by construction.
Thoughtful target-space engineering
Section 3.2 diagnoses a near-degenerate CineMA space (ED/ES cosine >0.99) and fixes it with a supervised 256-d phenotype bottleneck (LVEF, LVEDV, LV mass, RVEDV, LVGLS, AFib, MI); layer-9 selection (Section 4.3) is explicitly independent of Chagas labels.
Consistent gains over a strong baseline on large public data
Honest external benchmarking
Screening-relevant metric choice
Weaknesses
No inferential statistics
Missing alignment-method baselines
No ablation of the target space or objective
Overclaim in Discussion vs Table 3
Reproducibility and configuration opacity
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Scope & Claims
The paper (arXiv:2609.08582, cs.LG) asks whether structural knowledge distilled from cardiac MRI can improve ECG-only detection of Chagas cardiomyopathy — a disease whose patients and imaging are entirely absent from the pre-training corpus. The authors pre-train on 63,193 paired ECG-CMR UK Biobank exams, aligning ECG-FM's layer-9 representation to a CineMA-derived target via asymmetric InfoNCE (Section 3), then probe frozen for Chagas on CODE-15% + SaMi-Trop and enter the PhysioNet/CinC 2025 Challenge. Three claims are advanced (Section 1): the framework distills structural knowledge; it transfers to a disease never seen in pre-training; it is competitive on the official benchmark. All three receive evidence, of varying strength.
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 5 strengths & 5 weaknesses, analyzed in detail
- 5 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
Does the model need a cardiac MRI when diagnosing a patient?
No. Cardiac MRI is used only during pre-training on paired UK Biobank recordings; at deployment the system uses a standard 12-lead ECG alone, which is what makes it viable in resource-constrained settings.
How much does CMR alignment improve Chagas detection over an ECG-only model?
In five-fold cross-validation on CODE-15% and SaMi-Trop, AUROC improves from 0.827 to 0.851 and sensitivity within the top 5% of predicted risk improves from 0.377 to 0.427, compared with the same ECG foundation model without alignment.
Did this model win the PhysioNet 2025 Chagas Challenge?
No. Its overall challenge score of 0.269 ranked just below the top three teams (0.280-0.323), but it achieved the highest AUROC on SaMi-Trop-3 (0.773) and the best ELSA-Brasil challenge score (0.132) among those teams.
Were any Chagas patients used to train the representation?
No. Pre-training used 63,193 UK Biobank ECG-CMR pairs, which contain no Chagas cases; Chagas-labeled data (CODE-15%, SaMi-Trop) was used only afterward for the linear-probe evaluation.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- The provided text contains no appendix or supplementary material; if the arXiv PDF includes one, it was not part of this extraction.
- Table 1's AUROC column genuinely has no variance figures in the source text (verified present as printed); the missing SD is a property of the paper, not an extraction loss.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.