AI heart MRI measurements predict survival slightly better than clinicians in a 27,382-patient study
Researchers at City St George's, University of London and UCL report in The Lancet Digital Health that AI-derived ejection fraction was slightly but significantly more strongly linked to death risk than expert readings, at 2 to 3 seconds per scan.
Why it matters
Ejection fraction guides decisions on heart failure drugs, implanted devices, surgery and monitoring during chemotherapy, so a more consistent measurement can change who gets treated. The study followed patients in the UK and the United States for more than five years on average, during which 3,458 died. The researchers say clinicians should still check results, since AI errors affected about 1% of patients. Watch for prospective trials and whether hospitals adopt AI as the first-line read.
Key facts
- Patients: 27,382 in the UK and US
- Deaths during follow-up: 3,458
- AI analysis: about 2 to 3 seconds per patient
- Clinician analysis: around 13 minutes (earlier work)
- AI error rate: about 1% of patients, 0.08% of images
- Journal: The Lancet Digital Health