Source study found

Story checked

AI better assesses heart function from MRI than clinicians (opens in a new tab)

medicalxpress.com · 2026-11-06

Short answerEvidenceSource

Short answer

Mixed

Mixed.

2 claims go further than the study.

2 supported
2 overstated

Checked against the study summary. The full text wasn't available, so some details couldn't be settled either way.

Notice: Analysis is currently derived from study abstracts. Full-text paper analysis capabilities are currently in development.

Share this check

Help improve NewsLink

Tell us what worked or what needs attention in this check.

Follow the evidence trail

The story

AI better assesses heart function from MRI than clinicians

medicalxpress.com · 2026-11-06

The story’s checkable claims.

Read the original story

NewsLink checks it

Mixed

Two of four claims overstate the study. Two of four check out.

2 supported
2 overstated

Open claim evidence

The source study

Performance of artificial intelligence versus clinician left-ventricular segmentation of cardiac MRI for mortality prediction: a retrospective, multicentre, prognostic study

The Lancet. Digital Health · 2027

Open source record Open in Tessa

Then inspect each claim

Evidence layer

Claim by claim

Each claim gets a verdict. Expand it to see the evidence directly below.

Reading mode

Scan verdicts. Open evidence only when needed.

Browse by verdict

4 claims in this story

Showing all 4 claimsChoose a verdict to focus the list.

Claim 1 of 4OverstatedUsing artificial intelligence to assess heart function from MRI scans more strongly predicts survival than interpretation by clinicians, according to a study published online Nov. 2 in The Lancet Digital Health.View evidenceHide evidence

Why this verdict

The paper supports the underlying prognostic association: AI-derived CMR ejection fraction had marginally better all-cause mortality prediction than clinician-derived measurements. But the presented claim is marked as causal/headline-strength and says AI assessment more strongly predicts survival without conveying the retrospective observational design or small/marginal magnitude. The headline therefore outruns the abstract-level evidence and the story’s own more qualified body framing.

Study evidence

AI-derived ejection fraction provided marginally better prognostic performance for all-cause mortality than clinician-derived ejection fraction in a large retrospective multicentre CMR cohort.Unadjusted: AIC 57,936 vs 58,097 (ΔAIC = 161 favoring AI); C-index 0.669 vs 0.637 (ΔC = 0.033, 95% CI 0.026–0.041). Adjusted for age and sex: AIC 55,684 vs 55,742 (ΔAIC = 59); C-index 0.781 vs 0.777 (ΔC = 0.004, 95% CI 0.002–0.006).

“In this retrospective, multicentre, prognostic study, we studied a cohort of 27 382 consecutive patients who underwent cardiac MRI (CMR) in the UK and the USA.”

Claim 2 of 4OverstatedJoy said the findings suggest AI can measure this marker with remarkable consistency and precision, resulting in a small but meaningful improvement in identifying patients at higher risk.View evidenceHide evidence

As statedsmall but meaningful improvement

Why this verdict

The abstract supports a small/marginal prognostic improvement for AI-derived ejection fraction, but it does not establish that AI measured the marker with “remarkable consistency and precision,” nor does it prove clinical meaningfulness. The paper profile notes uncertainty about clinical impact and the need for prospective evaluation, so the quoted interpretation goes beyond the abstract-level evidence.

Study evidence

“In this retrospective, multicentre, prognostic study, we studied a cohort of 27 382 consecutive patients who underwent cardiac MRI (CMR) in the UK and the USA.”

Study evidence

Unadjusted model comparisons showed better overall model fit and discrimination for AI-derived ejection fraction versus clinician-derived ejection fraction (AIC 57,936 vs 58,097; ΔAIC = 161 favoring AI; C-index 0.669 vs 0.637; ΔC = 0.033, 95% CI 0.026–0.041).ΔAIC = 161; ΔC = 0.033 (95% CI 0.026–0.041)

“the incremental value of AI over clinician segmentation using likelihood-ratio tests of nested models.”

Claim 3 of 4SupportedGeorge Joy and colleagues assessed whether deep-learning AI segmentation of ejection fraction from cardiac MRI outperforms expert clinician segmentation for prediction of all-cause mortality in 27,382 consecutive patients with a baseline cardiac MRI.View evidenceHide evidence

As stated27,382 patients

Why this verdict

The abstract-level profile directly supports that Joy and colleagues studied deep-learning AI segmentation versus expert/routine clinician segmentation for all-cause mortality prediction in 27,382 consecutive baseline CMR patients in a retrospective multicentre cohort.

Study evidence

“In this retrospective, multicentre, prognostic study, we studied a cohort of 27 382 consecutive patients who underwent cardiac MRI (CMR) in the UK and the USA.”

Claim 4 of 4SupportedDuring 5.6 years of follow-up, AI ejection fraction showed better model fit (ΔAIC 161 favoring AI) and marginally better discrimination (ΔC index 0.033); after adjustment for age and sex, AI-derived ejection fraction still showed better model fit (ΔAIC 59) and marginally better discrimination (ΔC index 0.004).View evidenceHide evidence

As stated5.6 years of follow-up; ΔAIC 161; ΔC index 0.033; ΔAIC 59; ΔC index 0.004

Why this verdict

The numerical results match the profile: median follow-up 5.6 years; unadjusted ΔAIC 161 favoring AI and ΔC-index 0.033; after age/sex adjustment ΔAIC 59 and ΔC-index 0.004, described as marginally better discrimination.

Study evidence

“In this retrospective, multicentre, prognostic study, we studied a cohort of 27 382 consecutive patients who underwent cardiac MRI (CMR) in the UK and the USA.”

Study evidence

“the incremental value of AI over clinician segmentation using likelihood-ratio tests of nested models.”

Then look for missing context

Context layer

What the story left out

Important study details the story did not include.

Nested-model incremental-value testing found adding AI EF to clinician EF improved model fit substantially, whereas adding clinician EF to AI EF produced only a small improvement.
The paper profile treats this as a secondary contribution, but the presented story claims do not report the likelihood-ratio test results or the asymmetric incremental-value analysis in detail.
From retrospective multicentre prognostic observational study
Authors noted the need for prospective evaluation before broader clinical adoption or first-line use with clinician oversight.
The paper profile says prospective studies are needed before broader clinical adoption, but the story’s listed caveats do not mention this interpretation-changing limitation.
From Retrospective multicentre prognostic cohort; retrospective multicentre prognostic observational study

4 things the story did carry across

Large retrospective multicentre observational prognostic cohort of 27,382 consecutive cardiac MRI patients comparing AI-derived versus clinician-derived ejection fraction for all-cause mortality prediction.
AI-derived ejection fraction showed marginally better model fit and discrimination than clinician-derived ejection fraction, including after adjustment for age and sex.
Retrospective observational design limits causal interpretation.
Reported discrimination gains were numerically small, especially after age/sex adjustment, making clinical impact uncertain.

Then read the study layer

Study layer

Study at a glance

Scan the study first. Expand only the parts you want to inspect.

Pieces of work

Evidence read

study summary

Lead result

secondary data

1Lead resultsecondary dataCompare the prognostic performance (all-cause mortality prediction) of AI-derived versus clinician-derived left-ventricular volume metrics (especially ejection fraction) from cardiac MRI in a large retrospective, multicentre cohort.Retrospective multicentre prognostic cohortExpand

In plain English

Retrospective, multicentre prognostic cohort study (27,382 consecutive CMR patients from Barts Heart Centre and UPMC) comparing AI deep-learning versus routine clinician left-ventricular segmentation–derived ejection fraction for prediction of all-cause mortality using Cox proportional hazards models; AI-derived ejection fraction produced marginally better model fit (AIC) and discrimination (C-index) than clinician-derived ejection fraction, including after adjustment for age and sex.

Key findings

AI-derived ejection fraction provided marginally better prognostic performance for all-cause mortality than clinician-derived ejection fraction in a large retrospective multicentre CMR cohort.Unadjusted: AIC 57,936 vs 58,097 (ΔAIC = 161 favoring AI); C-index 0.669 vs 0.637 (ΔC = 0.033, 95% CI 0.026–0.041). Adjusted for age and sex: AIC 55,684 vs 55,742 (ΔAIC = 59); C-index 0.781 vs 0.777 (ΔC = 0.004, 95% CI 0.002–0.006).

“In this retrospective, multicentre, prognostic study, we studied a cohort of 27 382 consecutive patients who underwent cardiac MRI (CMR) in the UK and the USA.”

What this piece can’t prove

Retrospective, observational design (cohort assembled from existing reporting archives).
Reported improvements in discrimination are numerically small (described as marginal), so clinical impact is uncertain without further prospective study.

1 further detail could not be confirmed from the summary.

2secondary dataEstimate the incremental prognostic value of AI-derived ejection fraction over clinician-derived ejection fraction (and vice versa), including in models adjusted for age and sex.retrospective multicentre prognostic observational studyExpand

In plain English

The paper reports nested Cox model comparisons to quantify the incremental prognostic value of AI-derived left-ventricular ejection fraction (EF) versus clinician-derived EF for all-cause mortality. Using likelihood-ratio tests, AIC, and C-indices, the authors found that adding AI EF to clinician EF produced a substantial, statistically significant improvement in model fit, whereas adding clinician EF to AI produced only a small improvement; results were similar but attenuated after adjustment for age and sex.

Key findings

Unadjusted model comparisons showed better overall model fit and discrimination for AI-derived ejection fraction versus clinician-derived ejection fraction (AIC 57,936 vs 58,097; ΔAIC = 161 favoring AI; C-index 0.669 vs 0.637; ΔC = 0.033, 95% CI 0.026–0.041).ΔAIC = 161; ΔC = 0.033 (95% CI 0.026–0.041)
Incremental-value testing (likelihood-ratio tests) indicated that adding AI EF to a model containing clinician EF produced a large, statistically significant improvement (χ2 = 168, p < 0.0001), whereas adding clinician EF to a model containing AI EF produced only a small improvement (χ2 = 7.5, p = 0.023).LR test χ2 = 168 (p < 0.0001) for AI added to clinician; χ2 = 7.5 (p = 0.023) for clinician added to AI

“the incremental value of AI over clinician segmentation using likelihood-ratio tests of nested models.”

What this piece can’t prove

Retrospective, observational design (authors recommend prospective studies).
Abstract does not report detailed prespecification of nested-model comparisons or additional covariate handling beyond age and sex.

1 further detail could not be confirmed from the summary.

Finally, the search trail

Method layer