Abstract
Accuratehepaticvasculaturesegmentationiscriticalforsurgicalplanningandinterventionguidance,yetprogressisbottleneckedby annotation scarcity. High-quality hepatic vasculature annotation takes hours of expert effort per case. We present a physics-based synthetic data generation framework that produces realistic liver vasculature training data without manual annotation. Our framework combines space colonization for vessel synthesis withinpaintingandstyletransfertoproduceanatomicallyplausibleCTlabelpairs,relyingonlyonpredictionsfrombaselinesegmentationmodels. We show that synthetic data generated from predicted masks yields betterdownstreamperformancethanwhengeneratedfromexpertlabels. From 5 to 160 annotated cases, pretraining nnU-Net on synthetic data beforefinetuningonground-truth(GT)casesoutperformsGT-onlytrainingat nearlyeveryannotationbudget onbothbenchmarks, nearingthe performance of a GT-only baseline trained on all 252 cases. To the best of our knowledge, this is the first framework to generate realistic synthetic CT-vasculature pairs using only baseline-model predictions in place of expert annotation.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/SASHIMI_059.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=em0Pu6wQ1i
BibTex
@InProceedings{RisLeo_Realistic_MICCAISAT2026,
author = { Risteski, Leonid AND Fogarollo, Stefano AND Rauch, Nikolaus AND Harders, Matthias},
title = { { Realistic Synthetic Data for Annotation-Free Pretraining in Liver Vasculature Segmentation } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17258},
month = {pending},
page = {pending}
}