Health / Other
AI-derived heart MRI measure modestly improves mortality risk prediction
A retrospective comparison found a small advantage over clinician-derived measurements.
Status: source-checked.
An artificial intelligence method for measuring heart function from cardiac MRI scans modestly improved prediction of death compared with expert clinician measurements, according to a study reported Nov. 6 by Lori Solomon for Medical Xpress. The retrospective, multicenter research was published online Nov. 2 in The Lancet Digital Health.
George Joy of City St. George’s University of London and colleagues compared AI and clinician segmentation used to derive left-ventricular ejection fraction. The analysis included 27,382 consecutive patients with a baseline cardiac MRI and 5.6 years of follow-up. Its endpoint was prediction of death from any cause, not whether AI-guided treatment prolonged life.
The AI measure had a better statistical model fit and a slightly higher discrimination score than the clinician measure. Adding it to clinician-derived ejection fraction improved model fit more than adding the clinician measure to the AI result. The report describes a difference in prediction performance, not a test of replacing clinical judgment.
After adjustment for age and sex, the discrimination advantage narrowed to 0.004 on the C index, which the report characterized as marginal. That small difference limits how broadly the result should be interpreted. Whether use of the measure changes care or improves patient outcomes is not established by the retrospective prognostic comparison.
What we know / what we don't
What we know
- A retrospective comparison found a small advantage over clinician-derived measurements.
What we don't know
- Retrospective prognostic comparison, not evidence that AI-guided care improves survival or can replace clinicians.
The source record