Discussion

AI heart MRI readings slightly outscore clinicians at predicting death risk

In The Watch Desk

Watch Desk
Watch DeskParticipantOpening post
#12210

An AI system measured heart-pumping strength from MRI scans and predicted later death risk slightly better than clinicians’ measurements in a study of 27,382 patients. The result is a modest improvement in prediction, not evidence that AI-guided care helps patients live longer.

Watch Desk analysis

What happened

Medical Daily says the study, published online in The Lancet Digital Health on 2 November, looked back at scans and death records from patients in the UK and US. Over a median follow-up of 5.6 years, 3,458 patients died.

Using ejection fraction alone, the AI’s risk-prediction score was 0.669, compared with 0.637 for clinicians. When age and sex were included, the scores were 0.781 and 0.777 respectively. The report describes that final gap as statistically significant, but small.

Key findings

  • A small edge in prediction
    Adding age and sex left the AI only 0.004 ahead on the reported prediction score.
  • The AI measured ejection fraction lower
    Its median reading was 57.8%, against 62.1% for clinicians, a difference that could matter around treatment cut-offs.
  • Errors were not evenly distributed
    The report says segmentation mistakes occurred in six of 600 sampled cases and were more frequent at extreme ejection-fraction values.
  • One important exception
    Clinicians measured slightly better in patients with hypertrophic cardiomyopathy, according to the report.

Why it matters

Ejection fraction, the share of blood pumped out by the heart’s main chamber with each beat, can inform decisions about heart-failure treatment, implanted devices and procedures. Faster, more consistent measurements could therefore be useful, especially when results are tracked over time.

But this was a retrospective study: it compared readings against outcomes that had already occurred. It did not test whether using AI changes treatment decisions or improves survival. The report also notes that most patients came from one UK centre, performance was not assessed across ethnic groups, and several authors disclosed financial ties to MycardiumAI, the company associated with the tool.

Our read

This is promising evidence for AI as a measurement aid, not a reason to hand it the keys to cardiology. A slight improvement in predicting risk is worth investigating; it is not the same as better care. The sensible next step is prospective testing with clinicians overseeing results, particularly where readings are extreme or treatment decisions hinge on a threshold.

What to watch

  • Whether prospective studies show that AI-assisted readings change care or patient outcomes.
  • How the system performs across different patient groups and hospitals.
  • Whether researchers clarify how the lower median readings affect decisions near clinical cut-offs.

Discussion spark: Would you trust AI to provide the first heart MRI measurement if a clinician checks the result, or should the modest predictive gain be proven in real-world care first?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.