
PanAdipo AI Tool Proves Efficacy of AI-Enabled Cardiography for Characterizing EAT
Evangelos Oikonomou, MD, discusses the newly-developed deep-learning model and how it effectively characterizes epicardial adipose tissue and cardiovascular risk.
PanAdipo, a new view-agnostic video-based deep learning model developed for echocardiographic interpretation, has highlighted the efficacy of
These data were presented at the
“The focus of our work was to uncover a whole layer in routine cardiac imaging that we’ve just been ignoring, or perhaps just leaving on the table while interpreting those tests,” Oikonomou told HCPLive in an exclusive interview. “Whenever we order a cardiac ultrasound, we focus on the structure and anatomy of the heart – but we think there’s a whole additional layer where we can phenotype what we call adverse adiposity.”
Oikonomou and colleagues developed PanAdipo to characterize prominent epicardial adiposity from standard 2D B-mode echocardiographic acquisitions. It was tested across 5 major cohorts, which included a total of 1,114,441 videos from 28,797 random transthoracic echocardiographs (TTE) in 21,479 unique patients. The team analyzed discrimination of the report label and of CT-defined high indexed EAT volume, correlation with paired cardiac CT phenotypes, and body mass index (BMI)-adjusted associations with cardiometabolic biomarkers and newly documented metabolic disease.1,2
The 4 cohorts included temporally distinct Yale-New Haven Health System (YNHHS) TTEs (n = 4588), emergency department point-of-care ultrasound (POCUS) (n = 10,957), the community-based Multiethnic Study of Atherosclerosis (MESA) (n = 2740), and external MIMIC-IV TTE (n = 4549). In held-out TTE, PanAdipo successfully discriminated prominent epicardial adiposity with an area under the receiver operating characteristic curve (AUROC) of 0.91 (95% CI, 0.88-0.94), with saliency maps localizing attention to epicardial regions across views.1
Although the PanAdipo was modestly correlated with BMI (p = 0.19-0.4; P <.001), Oikonomou and colleagues saw a strong correlation with ectopic adiposity features on paired CT scans (n = 23), particularly pericardial fat fraction (p = 0.75) and skeletal muscle fat fraction (p = 0.74). In the MESA cohort, higher PanAdipo was also associated with higher homeostatic model assessment for insulin resistance (HOMA-IR) and triglyceride levels after adjustment for age, BMI, sex, and waist-hip ratio. Longitudinally, each 1-SD increment in log(PanAdipo) was also associated with a 21% and 37% higher risk of incident
Ultimately, Oikonomou and colleagues determined that the PanAdipo system successfully shifts routine echocardiography into a scalable AI-enabled biomarker for EAT. It also effectively captures cardiometabolic risk across community and clinical settings, supporting cardiac ultrasound as a method of metabolic risk stratification distinct from anthropometric indices.1
“The next step is to identify that optimal sweet spot where we’re really contributing value and we’re generating actionable information, which can inform clinical care, behavioral modifications for a patient, or perhaps even a more personalized allocation of emerging cardiometabolic therapies,” Oikonomou said. “Now, obviously, that would need to be studied in prospective clinical studies, but it could also become a kind of companion diagnostic for greater precision.”
Editors’ Note: Oikonomou reports no relevant disclosures.
References
Oikonomou E, Aminorroaya A, Abdollahi A, et al. PanAdipo: Turning Routine Echocardiography Into an AI-Enabled Epicardial Adiposity Biomarker. Presented at the American Heart Association (AHA) Scientific Sessions 2026, Chicago, IL. November 6-9, 2026.
Aminorroaya A, Coppi A, McNamara RL, et al. AI-Enabled Echocardiography Identifies Prominent Epicardial Adipose Tissue Associated With an Adverse Cardiometabolic Profile. Preprint. medRxiv. 2026;2026.07.22.26358690. Published October 9, 2026. Accessed November 17, 2026.
doi:10.64898/2026.07.22.26358690
Related to this article








