Journal of the American College of Cardiology

Heart ultrasound model flags fat linked to metabolic risk

The model scored 0.86 for distinguishing scans labeled as showing prominent fat around the heart, on a statistical measure of how well it separated groups.

Figure 1 (preprint). PanAdipo attends to the epicardial area, discriminates across views, and is only weakly correlated with conventional echocardiographic features.
Open the figure at full size
Figure 1 (preprint). PanAdipo attends to the epicardial area, discriminates across views, and is only weakly correlated with conventional echocardiographic features.PanAdipo attends to the epicardial area, discriminates across views, and is only weakly correlated with conventional echocardiographic features.Aminorroaya et al.

Our read

Biomarkers
Studied:
Cross-sectional study of 950 middle-aged to older adults in MESA
What changed:
Higher ultrasound model scores were linked to greater insulin resistance after accounting for age, sex, and body size
The catch:
Training labeled undocumented epicardial fat as absent, potentially misclassifying patients
In the record:
One of 3 papers on PanAdipo on record: 3 with results in people.

Editor of record: Boris Djordjevic · AgingResearch.News

Follow @agingbrief

Heart ultrasound scans from patients in the Yale-New Haven Health System formed the basis of an observational study of fat around the heart. Researchers developed PanAdipo, an artificial intelligence model, using videos from 28,797 scan studies. They evaluated it in five testing cohorts, including 5,167 patients with paired ultrasound and computed tomography scans—images made using X-rays—taken within 12 months.

The model distinguished unusually high fat volume relative to heart size on computed tomography scans, scoring 0.90 on the discrimination measure. Among 950 participants in the community-based Multi-Ethnic Study of Atherosclerosis, higher scores were associated with greater insulin resistance—a poorer response to insulin—after accounting for age, sex and body mass index.

Why it matters

Metabolic risk assessment bears on diseases common in later life. The authors suggest that heart ultrasound could offer a scalable way to assess that risk.

Caveats

The model learned from clinicians’ report labels rather than direct measurements of heart fat. Health differences not captured by age, sex and body mass index may contribute to the association with insulin resistance.

From the paper

Participants

Across the four testing cohorts and the held-out CT-linked cohort, we analyzed 4,588 individuals in the YNHHS TTE held-out test set, 10,957 in YNHHS POCUS, 4,549 in MIMIC-IV, 2,740 in MESA, and 5,167 in the YNHHS TTE-CCTA cohort (, Supplemental Figure 1).

Results, Cohort composition

Result

… the associations with hyperlipidemia (HR 1.11, 95% CI 1.04-1.18) and MASLD/MASH (HR 1.31, 95% CI 1.03-1.67) persisted, including with multiple imputation of missing BMI and in Fine-Gray models, whereas those with T2D and hypertension did not (Supplemental Table 10).

Results, PanAdipo is associated with newly documented cardiometabolic disease in health systems

Limitation

First, the training label was an expert-reader assignment of a prominent epicardial fat depot extracted from routine clinical reports, which is more specific than sensitive.

Study Limitations

Quoted word for word from the full text on medRxiv.

Comments

Be the first to comment.