AI-Enabled Echocardiography Identifies an Adverse Epicardial Adiposity Phenotype Associated with Cardiometabolic...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Excess epicardial adipose tissue (EAT) is linked to cardiovascular-kidney-metabolic (CKM) dysfunction.
• Traditional assessment of EAT requires advanced imaging techniques.
• Scalable phenotyping of EAT is needed for cardiometabolic risk stratification.

Purpose of the Study:

• To evaluate AI-enhanced echocardiography for assessing epicardial adiposity.
• To determine if AI can identify individuals at increased cardiometabolic risk.
• To develop a scalable method for EAT assessment.

Main Methods:

• A deep learning model, PanAdipo, was developed using over 1.1 million echocardiography videos.
• The model was validated in four diverse cohorts, including a large health system, an emergency department cohort, MIMIC-IV, and the MESA study.
• Analyses included discrimination of EAT, independence from conventional echocardiography, spatial explainability, correlation with CT imaging, and association with cardiometabolic biomarkers and disease.

Main Results:

• PanAdipo accurately discriminated prominent EAT (AUROC 0.91) and localized attention to the epicardial area.
• The AI score strongly correlated with epicardial adiposity on CT scans (Spearman ρ=0.75).
• Higher PanAdipo scores were independently associated with metabolic dysfunction (e.g., HOMA-IR, triglycerides) and incident metabolic disease, even after adjusting for BMI.

Conclusions:

• AI-enhanced echocardiography offers a scalable, view-agnostic biomarker for adverse epicardial adiposity.
• This technology can characterize cardiometabolic dysfunction.
• Echocardiography, augmented by AI, can play a new role in CKM risk stratification.