Synthetic Echocardiograms from Diffusion Models in Rare Cardiovascular Disease
- 1Duke University
Abstract
The limited availability of imaging data for uncommon cardiovascular phenotypes constrains the development of robust imaging models. We evaluated whether class-conditional diffusion models can generate synthetic transthoracic echocardiograms that improve downstream cardiac imaging tasks. The primary application was cardiac amyloidosis detection in a Duke University cohort using a two-step classifier, with external analyses using EchoNet-Dynamic for image-fidelity assessment and EchoNet-LVH for wall-thickness phenotype classification. The Duke cohort was partitioned at the patient-encounter level into 70% training, 15% validation, and 15% test sets; generators were trained only on the training partition, augmentation levels were selected using validation AUROC, and final evaluation used held-out real test data. In the Duke all-view two-step analysis, adding synthetic images increased AUROC from 0.883 to 0.924, with an AUROC difference of 0.041 (95% CI, 0.013-0.069); in EchoNet-LVH, AUROC increased from 0.832 to 0.864, with an AUROC difference of 0.032 (95% CI, 0.021-0.044). Expert review found that synthetic images were sometimes difficult to identify as synthetic, but rated them lower for diagnostic adequacy. These findings suggest that diffusion-generated echocardiograms may provide a practical approach to augmenting limited training data for selected cardiac imaging tasks and motivate further evaluation across clinical settings.
Related
- Cardiovascular Risk of Conventional Synthetic, Biologic and Targeted Synthetic Disease-Modifying Antirheumatic Drugs in Common Rheumatic Diseases: A Narrative Review of Current Evidence.
- Diffusion-weighted magnetic resonance imaging and immunohistochemistry in Epstein-Barr virus-positive diffuse large B-cell lymphoma of the maxilla: a rare case report
- A physiologically constrained calibration framework for cardiovascular models applied to a paediatric synthetic sepsis population
- Texture-aware diffusion modelling for severity-conditioned synthetic progression-like visualization of diabetic foot ulcers from single images
- Effects of a Program to Reduce Unnecessary Echocardiograms and Nuclear Stress Tests for Cardiopulmonary Complaints in an Emergency Department-Run Observation Unit
BibTeX
@misc{lyu2026synthetic,
title = {{Synthetic Echocardiograms from Diffusion Models in Rare Cardiovascular Disease}},
author = {P. Lyu and R. Henao and L. C. Kwee and F. Z. Peng and S. Vemulapalli and S. H. Shah and M. G. Khouri and A. Zhang},
howpublished = {medRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.14.26363024},
url = {https://www.medrxiv.org/content/10.64898/2026.09.14.26363024v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - Synthetic Echocardiograms from Diffusion Models in Rare Cardiovascular Disease
AU - P. Lyu
AU - R. Henao
AU - L. C. Kwee
AU - F. Z. Peng
AU - S. Vemulapalli
AU - S. H. Shah
AU - M. G. Khouri
AU - A. Zhang
PY - 2026
DA - 2026/09/14
PB - medRxiv
DO - 10.64898/2026.09.14.26363024
UR - https://www.medrxiv.org/content/10.64898/2026.09.14.26363024v1
AB - The limited availability of imaging data for uncommon cardiovascular phenotypes constrains the development of robust imaging models. We evaluated whether class-conditional diffusion models can generate synthetic transthoracic echocardiograms that improve downstream cardiac imaging tasks. The primary application was cardiac amyloidosis detection in a Duke University cohort using a two-step classifier, with external analyses using EchoNet-Dynamic for image-fidelity assessment and EchoNet-LVH for wall-thickness phenotype classification. The Duke cohort was partitioned at the patient-encounter level into 70% training, 15% validation, and 15% test sets; generators were trained only on the training partition, augmentation levels were selected using validation AUROC, and final evaluation used held-out real test data. In the Duke all-view two-step analysis, adding synthetic images increased AUROC from 0.883 to 0.924, with an AUROC difference of 0.041 (95% CI, 0.013-0.069); in EchoNet-LVH, AUROC increased from 0.832 to 0.864, with an AUROC difference of 0.032 (95% CI, 0.021-0.044). Expert review found that synthetic images were sometimes difficult to identify as synthetic, but rated them lower for diagnostic adequacy. These findings suggest that diffusion-generated echocardiograms may provide a practical approach to augmenting limited training data for selected cardiac imaging tasks and motivate further evaluation across clinical settings.
ER -