AHA Scientific Sessions 2026 AbstractNov 9, 2026, 00:39 GMT

AI detects prominent fat around the heart in ultrasound videos

The model scored 0.91 on a statistical measure of its ability to distinguish prominent fat around the heart in separate test data.

Paper by Evangelos K. Oikonomou, Arya Aminorroaya, Ashkan Abdollahi, +4Rohan Khera · Yale University

A heart cross-section surrounded by thick fat sits beside a smaller heart with a thin fat layer and a liver dotted with fatty patches.

Our read

Biomarkers

Studied:: Cohort study of 4,588 patients undergoing echocardiography (Independent replication)
What changed:: PanAdipo discriminated prominent epicardial adiposity with an AUROC of 0.91 (95% CI, 0.88-0.94) in held-out TTE
The catch:: Comparison with detailed fat measurements involved only 23 paired scans
In the record:: One of 3 papers on PanAdipo on record: 3 with results in people.

Editor of record: Boris Djordjevic · AgingResearch.News

Paper

Heart ultrasound videos from 21,479 Yale-New Haven Health System patients supplied training data for PanAdipo, an artificial intelligence model, in an observational study. Researchers tested it in four independent groups: later Yale-New Haven scans (4,588; age 66; 50.1% female), emergency-department scans (10,957; age 66; 46.6% female), community participants (2,740; age 74; 52.8% female), and an external ultrasound dataset (4,549; age 70; 51.4% female).

In follow-up analyses, each one-standard-deviation increase in the log-transformed score was associated with 21% higher relative risk of type 2 diabetes and 37% higher relative risk of fatty liver disease linked to metabolic dysfunction. Standard deviation measures variation; a logarithm changes the score’s scale. The score correlated only modestly with body mass index, a measure relating weight to height.

Why it matters

Metabolic diseases can complicate healthy aging. The authors propose using routine heart ultrasound to assess metabolic risk beyond body-size measures.

Caveats

Other differences in metabolic health could help explain the disease links. The comparison with detailed fat measurements involved only 23 paired scans, and the findings were reported in a conference abstract.