Chest radiograph model for prescreening reduced ejection fraction, tested in three Korean cohorts
DLLV, a deep learning model that flags a left ventricular ejection fraction of 40% or less on chest radiographs, reached AUCs of 0.86 in a tertiary-centre cohort and 0.92 to 0.97 in two health-checkup cohorts, and as a prescreen was estimated to cut echocardiography volume by up to 85% without missing any of the 19 affected people in those cohorts.
- Model
- DLLV
- Field
- Medicine
- Date
- 2026-10-21
- Human collaborators
- Sowon Jang, Ju Gang Nam
What was found
The model was developed on 24,674 patients with 32,135 same-day radiograph and echocardiogram pairs from 2016 to 2021, using counterfactual learning to improve performance and the localisation of its heatmaps. External testing covered a tertiary referral cohort of 5,440 patients and two health-checkup cohorts of 28,034 and 6,085. In the checkup cohorts the area under the precision-recall curve was 0.04, which still corresponds to 87-fold and 37-fold enrichment over the low baseline prevalence.
Novelty check
Europe PMC, 2026-10-29: chest radiograph or chest x-ray with ejection fraction and deep learning or artificial intelligence, in title or abstract, returned 6 records, including a 2023 Lancet Digital Health multi-institutional model classifying cardiac function from chest radiographs and a 2022 Canadian Journal of Cardiology model detecting left ventricular systolic dysfunction. Detecting reduced ejection fraction on radiographs is not new; this adds a prescreening analysis in health-checkup populations.
Caveats and known objections
Retrospective, developed at Seoul National University hospitals and tested in Korean cohorts, with one co-author from the AI department of Osstem Implant and no funding stated in the abstract; the external cohorts are separate datasets, but the testing was the developers' own. Positive predictive value is low by design: an AUPRC of 0.04 means most flagged people would not have reduced ejection fraction, and the no-missed-cases figure rests on 19 affected people. The endpoint is detection against echocardiography, a surrogate. Autonomy is ai-assisted: the model flags candidates for an echocardiogram that clinicians then order and read.
No independent check is recorded on this entry yet.
Reading the primary source closely enough to say whether it supports the claim counts as a check, and you are credited on the entry.
Or on GitHub: submit a check challenge the grade send a correction or send a pull request
Flag this for triage
Signals order the review queue and nothing else. They are never published, and they never move a grade: that takes a citation.
Entry history (1 event)
- AddedEntered the registry graded Peer reviewed and AI-assisted.
Entries are never deleted. A grade that does not hold up is downgraded on the record, with the reason beside it.
Graded peer reviewed for verification and ai-assisted for autonomy. What these mean.
Cite this entry
whataifound.org. (2026). Chest radiograph model for prescreening reduced ejection fraction, tested in three Korean cohorts. whataifound.org: A Registry of AI Scientific and Mathematical Discoveries. https://whataifound.org/finding/2026-10-21-dllv-chest-radiograph-lv-dysfunctionBibTeX
@misc{whataifound-seoulnationaluniversitybundanghospital-2026-dysfunction,
title = {Chest radiograph model for prescreening reduced ejection fraction, tested in three Korean cohorts},
author = {{whataifound.org}},
year = {2026},
howpublished = {whataifound.org: A Registry of AI Scientific and Mathematical Discoveries},
note = {Result by Seoul National University Bundang Hospital. Verification: Peer reviewed. Autonomy: AI-assisted.},
url = {https://whataifound.org/finding/2026-10-21-dllv-chest-radiograph-lv-dysfunction}
}