A Foundational EDM2-Based Generative Model for High-Resolution Synthetic Fetal Ultrasound Imaging from Open Datasets

2026-08-05

AI Summary

This study addresses the challenges in fetal ultrasound AI research—namely data scarcity, privacy constraints, and annotation difficulties—stemming from the absence of high-quality public datasets. To overcome these limitations, the authors propose the first application of the EDM2 diffusion model to synthesize fetal ultrasound images, integrating multiple publicly available data sources to generate high-fidelity 512×512 images across six anatomical classes. Without relying on large-scale private datasets, the method achieves substantially improved image quality, as evidenced by lower FID scores, and enhances downstream task performance: fine-tuning on the synthetic data yields a classification ensemble accuracy of 93.36%, surpassing baselines trained solely on real data. Clinical expert evaluation assigns the generated images a realism score of 2.67 out of 5, indicating moderate fidelity and potential clinical utility.

Abstract

Prenatal ultrasound imaging is key for assessing fetal health, but AI progress is limited by scarce, privacy-restricted, and hard-to-annotate datasets. We propose a high-resolution fetal ultrasound synthesis framework based on the EDM2 diffusion architecture, trained on multiple public datasets to generate 512x512 images across six anatomical classes. Our method achieved improved image quality with lower FID scores and enhanced downstream fetal plane classification, reaching 93.36% ensemble accuracy after fine-tuning, surpassing real-data-only training. Clinical evaluation by an experienced fetal ultrasound specialist (10+ years) on 100 images yielded a mean realism score of 2.67/5, with real images rated higher than synthetic. Artefacts included smoothing, speckle irregularities, and anatomical inconsistencies. Code, data, models and other resources to reproduce this work are available at https://github.com/xfetus/fetal-ultrasound-edm2.

Problem

• fetal ultrasound
• data scarcity
• privacy constraints
• medical image synthesis
• annotation difficulty

Innovation

• EDM2
• diffusion model
• synthetic fetal ultrasound
• high-resolution image generation
• medical image synthesis