PanAdipo

Epicardial adiposity from any routine echocardiogram

PanAdipo is a video model that reads standard 2D echo clips (any view, TTE or POCUS), scores prominent epicardial fat, and maps the score to CT-measured epicardial adipose tissue (EAT), with an age- and sex-specific nomogram.

Central illustration: PanAdipo development and evaluation cohorts, and findings for CT-quantified epicardial fat, metabolic profile in MESA, and newly documented metabolic disease
Study overview (central illustration).

How it works

The model runs on your own computer: images never leave it. This page only does arithmetic on scores, so you can explore the nomogram and review results without any data being sent anywhere.

Step 1

Load a DICOM study

Cine clips are masked to the ultrasound sector (no text, ECG or calipers) and resized to 256 × 256.

Step 2

Score every clip

PanAdipo scores 16 frames per clip; the study score is the mean over clips.

Step 3

Estimate EAT

The score maps to CT EAT volume (mL) and EAT / heart volume, calibrated on paired CCTA.

Step 4

Nomogram & tracking

Age and sex give the probability of high EAT and a screening threshold; BMI flags occult adiposity.

Echo frame, PanAdipo Grad-CAM++ attention over the epicardium, and paired CT segmentation of cardiac fat and heart volume
Grad-CAM++ attention over the epicardium (left) and the paired CCTA reference standard (right): segmentation of epicardial and paracardial fat and of the cardiac structures summed as heart volume.

Nomogram calculator

Enter a PanAdipo study score (from the pipeline) with the patient's age and sex. BMI is optional and enables the occult-adiposity flag.

Slider is on a log scale (0.001 – 1).
Probability of high EAT
High EAT: EAT / heart volume ≥ 95th percentile for age group and sex (CT reference).

Track results over time

Open a study_predictions.csv produced by the pipeline (or several). The file is read by your browser only. Patients with two or more dated studies can be followed over time; occult adiposity and screen-positive studies are summarised.

…or drop CSV files here. Needs study_id, panadipo_score; uses patient_id, study_date, age, sex, bmi when present.

Running PanAdipo

The model runs locally on your own computer (ideally with a GPU), so images never leave your institution. It produces the study_predictions.csv used by the tracker above. For access to the pipeline and model weights, contact evangelos.oikonomou@yale.edu.

Cite

Aminorroaya A, …, Oikonomou EK*. AI-Enabled Echocardiography Identifies Prominent Epicardial Adiposity Associated with an Adverse Cardiometabolic Profile. Under review, 2026. *Corresponding author: evangelos.oikonomou@yale.edu