AI against clinician left-ventricular segmentation of cardiac MRI for predicting mortality in 27,382 patients
In 27,382 patients scanned at Barts Heart Centre in London and the University of Pittsburgh Medical Center and followed for a median 5.6 years, ejection fraction measured by deep-learning segmentation of cardiac MRI predicted death slightly better than the clinicians' own measurements (C index 0.669 against 0.637), a gap that shrank to 0.781 against 0.777 after adjustment for age and sex.
- Model
- Deep-learning left-ventricular segmentation of cardiac MRI, not named in the abstract
- Field
- Medicine
- Date
- 2026-10-01
- Human collaborators
- James C. Moon, Rhodri H. Davies
What was found
The cohort combined 26,106 consecutive patients scanned at Barts Heart Centre between October 2014 and December 2019, referred from nine hospitals, with 1,276 scanned at UPMC between June 2010 and March 2016; 3,458 (13%) died during follow-up. AI segmentation was run on every scan and Cox models compared the prognostic value of AI-measured and clinician-measured left-ventricular volumes. AI ejection fraction gave better model fit (a difference in AIC of 161) and slightly better discrimination (a difference in C index of 0.033, 95% CI 0.026-0.041), and adding it to the clinician measurement improved fit. The authors conclude that AI could be used for first-line measurement of ejection fraction with clinician oversight, and that this should be tested prospectively.
Novelty check
Europe PMC, 2026-11-16, cardiac magnetic resonance or cardiac MRI with artificial intelligence or deep learning, ejection fraction, mortality and segmentation: 817 records, mostly segmentation-accuracy and imaging-review papers; among the top results this was the only head-to-head comparison against clinicians on mortality. Nothing in this registry covered cardiac MRI.
Caveats and known objections
Retrospective, with both measurements taken from the same archived scans, so it shows which number predicts death better, not what happens when clinicians use the AI. The advantage is marginal and almost disappears after adjustment for age and sex. The abstract does not name the segmentation model or say who developed it, so whether this is an external validation by unaffiliated investigators cannot be told from it. The UPMC cohort is small relative to the Barts one. Autonomy is ai-assisted, as for other measurement studies in this registry: the model measures and no care decision was studied, and the authors recommend clinician oversight.
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). AI against clinician left-ventricular segmentation of cardiac MRI for predicting mortality in 27,382 patients. whataifound.org: A Registry of AI Scientific and Mathematical Discoveries. https://whataifound.org/finding/2026-10-01-ai-cmr-ejection-fraction-mortality
BibTeX
@misc{whataifound-universitycollegelondon-2026-mortality,
title = {AI against clinician left-ventricular segmentation of cardiac MRI for predicting mortality in 27,382 patients},
author = {{whataifound.org}},
year = {2026},
howpublished = {whataifound.org: A Registry of AI Scientific and Mathematical Discoveries},
note = {Result by University College London. Verification: Peer reviewed. Autonomy: AI-assisted.},
url = {https://whataifound.org/finding/2026-10-01-ai-cmr-ejection-fraction-mortality}
}