Abstract
Magnetic Resonance Imaging (MRI) plays a crucial role in the investigation of the uterus, allowing for detailed visualization and segmentation of anatomical structures. In this work, we tested interactive segmentation foundation models for the segmentation of the uterine wall and cavity. We apply and evaluate VoxTell, MedSAM2D, and nnInteractive using text-, box-, and point-based prompts, and compare them to nnU-Net. For repeatability and reproducibility, we use the Ground Truth (GT) for prompt creation and propose a five-point strategy for multi-point prompting. Our results show that nnU-Net outperforms the interactive foundation models in terms of Dice similarity coefficient and Hausdorff distance. Additional analyses show that performance decreases with lesion volume, distance from the prompt, and spatial resolution. We further analyze the behavior of bounding box and multi-point prompts, showing that the proposed sampling strategy outperforms random point selection. Overall, our results highlight key limitations of current interactive foundation models and identify factors that make uterine segmentation particularly challenging.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CAPI_WOMEN_007.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=Ss3kFaUK1p
BibTex
@InProceedings{GraCle_Interactive_MICCAISAT2026,
author = { Grange, Clemens AND Fischer, Stefan M. AND Reithmeir, Anna AND Schnabel, Julia A. AND Felsner, Lina},
title = { { Interactive Foundation Models for 3D MR Uterus Segmentation: Performance and Limitations } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17256},
month = {pending},
page = {pending}
}