Pathological MRI segmentation by synthetic pathological data generation in fetuses and neonates

Abstract Background Automated analysis of fetal and neonatal brain MRI is limited by the scarcity of annotated pathological datasets and by privacy constraints that can restrict data sharing, particularly for rare conditions. This study investigated whether synthetic pathological fetal and neonatal brain MRI generated from semantic label images can improve segmentation performance, particularly in abnormal developing brains. Methods In this work, we introduce Fetal&Neonatal-DDPM, a novel 3D conditional diffusion model framework designed to generate high-quality synthetic pathological fetal and neonatal MRIs from semantic label images. Further, we enhance training data by modifying healthy label images through morphological alterations to simulate conditions such as ventriculomegaly, cerebellar and pontocerebellar hypoplasia, and microcephaly. Synthetic images were evaluated by four board-certified radiologists using a four-level image-quality and diagnostic-usability score. Segmentation utility was assessed by training nnU-Net models with different combinations of real and synthetic data and evaluating Dice scores on the FeTA2021 MICCAI challenge test set and a clinical cohort of severe spina bifida-associated ventriculomegaly. Results Fetal&Neonatal-DDPM generated visually realistic fetal and neonatal brain MRIs from original and pathologically modified label images. Synthetic images received significantly higher average radiological quality and diagnostic-usability scores than the sampled real images. Synthetic pathological data enhanced state-of-the-art nnUNet segmentation performance, particularly for severe ventriculomegaly cases, with the greatest significant ( p < 0.05) improvements achieved in ventricle segmentation (Dice scores: 0.9290 vs. 0.7317). Conclusions This study underscores the potential of generative AI as transformative tool for data augmentation, offering improved segmentation performance in pathological cases. This development represents a significant step towards improving analysis and segmentation accuracy in prenatal imaging, and also offers new ways for data anonymization through the generation of pathologic image data. Our Fetal&Neonatal-DDPM code, label modification framework, and fetal examples for pathological fetal MRI synthesis are available at https://github.com/Mishakaandorp/Fetal_neonatal_DDPM .

Authors

Institutions

Publication Details

Journal
BMC Medical Imaging
Published
2026-11-16
DOI
https://doi.org/10.1186/s12880-026-03027-2
Citations
8
Primary Topic
Medical Image Segmentation Techniques
Type
article
Field-Weighted Citation Impact
6.84

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Pathological MRI segmentation by synthetic pathological data generation in fetuses and neonates

Patrice Grehten, Misha Kaandorp, Hosna Asma-ull, Levente Lánczi et al.
8 citations
BMC Medical Imaging
Medical Image Segmentation Techniques
6.84
article

Pathological MRI segmentation by synthetic pathological data generation in fetuses and neonates

Patrice Grehten, Misha Kaandorp, Hosna Asma-ull, Levente Lánczi, András Jakab, Damola Agbelese, Hyun Gi Kim, Kelly Payette, Antonio Giulio Gennari
article en
8 citations

Abstract

Abstract Background Automated analysis of fetal and neonatal brain MRI is limited by the scarcity of annotated pathological datasets and by privacy constraints that can restrict data sharing, particularly for rare conditions. This study investigated whether synthetic pathological fetal and neonatal brain MRI generated from semantic label images can improve segmentation performance, particularly in abnormal developing brains. Methods In this work, we introduce Fetal&Neonatal-DDPM, a novel 3D conditional diffusion model framework designed to generate high-quality synthetic pathological fetal and neonatal MRIs from semantic label images. Further, we enhance training data by modifying healthy label images through morphological alterations to simulate conditions such as ventriculomegaly, cerebellar and pontocerebellar hypoplasia, and microcephaly. Synthetic images were evaluated by four board-certified radiologists using a four-level image-quality and diagnostic-usability score. Segmentation utility was assessed by training nnU-Net models with different combinations of real and synthetic data and evaluating Dice scores on the FeTA2021 MICCAI challenge test set and a clinical cohort of severe spina bifida-associated ventriculomegaly. Results Fetal&Neonatal-DDPM generated visually realistic fetal and neonatal brain MRIs from original and pathologically modified label images. Synthetic images received significantly higher average radiological quality and diagnostic-usability scores than the sampled real images. Synthetic pathological data enhanced state-of-the-art nnUNet segmentation performance, particularly for severe ventriculomegaly cases, with the greatest significant ( p < 0.05) improvements achieved in ventricle segmentation (Dice scores: 0.9290 vs. 0.7317). Conclusions This study underscores the potential of generative AI as transformative tool for data augmentation, offering improved segmentation performance in pathological cases. This development represents a significant step towards improving analysis and segmentation accuracy in prenatal imaging, and also offers new ways for data anonymization through the generation of pathologic image data. Our Fetal&Neonatal-DDPM code, label modification framework, and fetal examples for pathological fetal MRI synthesis are available at https://github.com/Mishakaandorp/Fetal_neonatal_DDPM .

BMC Medical Imaging
University of Debrecen (HU), King's College London (GB), University of Zurich (CH), Asan Medical Center (KR), University of Ulsan (KR), University Children's Hospital Zurich (CH), Catholic University of Korea (KR)
National Science Foundation, Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung
Good health and well-being
Openalex Percentile: Top 3%
Medical Image Segmentation Techniques
6.84
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon