Cross-Modal MRI Ovary Segmentation in Endometriosis Using Unpaired TVUS Prototype Priors

Xingjian Kang; Lina Felsner; Dominik Perrin; Daiqi Liu; Jasmin Arjomandi; Franziska Mathis-Ullrich; Alexandra Stoll; Katharina Breininger

cs.CVAug 19, 2026

Cross-Modal MRI Ovary Segmentation in Endometriosis Using Unpaired TVUS Prototype Priors

Authors: Xingjian Kang, Lina Felsner, Dominik Perrin, Daiqi Liu, Jasmin Arjomandi, Franziska Mathis-Ullrich, Alexandra Stoll, Katharina Breininger

Organizations: Center for AI and Data Science (CAIDAS), Julius-Maximilians-Universität Würzburg, Würzburg, Germany · Institute for Computational Imaging and AI in Medicine (CompAI), Technical University of Munich, Munich, Germany · Pattern Recognition Lab, Friedrich-Alexander Universität Erlangen-Nürnberg, Erlangen, Germany · Department Artificial Intelligence in Biomedical Engineering (AIBE), Friedrich-Alexander-Universität Erlangen-Nürnberg, Erlangen, Germany

Abstract

Transvaginal ultrasound (TVUS) and magnetic resonance imaging (MRI) provide complementary information for endometriosis image analysis, yet existing studies mainly focus on single-modality analysis or disease classification, leaving cross-modal ovarian segmentation largely unexplored. In this work, to tackle the increased difficulty of ovary segmentation in MRI due to ovaries' small target size and ambiguous boundaries with surrounding pelvic structures, we propose a dual branch framework for ovary segmentation across TVUS and MRI. More specifically, by adapting MedSAM3 with TVUS-derived prototype bank, we aim to align anatomically consistent feature representations across both modalities. Extensive experiments are conducted on endometriosis-related TVUS and MRI datasets. We observe quantitative and qualitative improvements of over 5 percentage points for the proposed dual-branch approach compared with multiple state-of-the-art methods. Furthermore, our ablation study shows the contribution of individual components such as the prototype bank and the importance of warm-up pretraining in the source TVUS domain.

Figures & tables

Figure 1: Overview of our cross-modal prototype-contrastive learning framework. Unpaired TVUS and MRI images are encoded in a shared representation space, where prototype contrastive learning promotes cross-modal anatomical consistency.

Figure 2: Performance comparison between proposed methods and baselines on MRI ovary segmentation. The best performance is bold. “FT” refers to the fully supervised LoRA finetuning on the MRI dataset. “Ours” denotes the proposed dual-branch approach with a MedSAM3 backbone.

Panel of Figure 3 — Figure 3: Qualitative comparison of ovary segmentation results on MRI. Each row corresponds to one testing case, and the columns show the ground truth (GT) and segmentation results from different methods. "FT" denotes fine-tuning. The red dashed box indicates the valid evaluation region, with predictions outside it ignored during training and evaluation.

Warm-up phase			With
TVUS	MRI	Epoch(s)	Bank	mDice	mAP
✓		3	✓	61.02 $\pm$ 3.40	27.28 $\pm$ 0.98
✓		3		57.74 $\pm$ 4.19	28.37 $\pm$ 1.52
✓	✓	3	✓	17.92 $\pm$ 2.46	6.45 $\pm$ 2.25
		0	✓	45.96 $\pm$ 5.49	24.92 $\pm$ 1.58
✓		5	✓	50.86 $\pm$ 6.70	27.63 $\pm$ 0.53

Table 1: Performance comparison about different prototype bank and warm-up setting. First 2 columns indicate which branch has been activated in the warm-up phase. All of the experiments use MedSAM3 as backbone model. The best performance of each metric is in bold , second best underlined .

$\lambda$			$\alpha$
Value	mDICE	mAP	Value	mDICE	mAP
0.005	$58.82\pm 5.22$	$27.19\pm 0.84$	0.75	$57.96\pm 11.14$	$26.61\pm 1.64$
0.010	$57.23\pm 7.99$	$26.09\pm 1.31$	0.80	$59.42\pm 5.36$	$27.77\pm 0.81$
0.015	$\mathbf{61.02\pm 3.40}$	$\mathbf{27.28\pm 0.98}$	0.85	$59.56\pm 9.03$	$25.13\pm 1.43$
0.020	$59.07\pm 5.24$	$26.70\pm 1.77$	0.90	$59.31\pm 7.21$	$\mathbf{28.04\pm 1.27}$
0.025	$56.51\pm 10.37$	$26.35\pm 2.86$	0.95	$\mathbf{61.02\pm 3.40}$	$27.28\pm 0.98$

Table 2: Sensitivity analysis about key parameters.

\lambda

indicates the contrastive loss weight, while

\alpha

denotes the EMA momentum for updating the TVUS prototype. The best performance of each metric is in bold .

Cross-Modality Medical Image Segmentation

Cross-modality medical image segmentation labels anatomy and disease across imaging types despite differences in appearance, resolution, and available annotations.

See all 72 papers

Topology-Informed Prompt-Conditioned Universal Segmentation of Uterine Structures from Ultrasound and MRI
Oct 5, 2026Yongheng Sun, Yuexi Gu, Jingwen Sun +2Ultrasound Image Segmentation Cross-Modality Medical Image Segmentation
Department of Radiology and BRIC, UNC at Chapel Hill, Chapel Hill, NC 27599, USA
OvAi Focus: AI-based Multi-class Segmentation of Functional Ovaries and Adnexal Masses in Gynecological Ultrasound
Jul 15, 2026Niccolò Tallone, Francesca Salis, Pio Raffaele Fina +13Ultrasound Image Segmentation Medical Image Segmentation
SynDiag s.r.l., Turin, Italy · Academic Division of Gynecology and Obstetrics, University of Turin, Azienda Ospedaliera Ordine Mauriziano, Turin, Italy · University of Turin +5
Unified CT and MRI Pancreas Segmentation for Label-Efficient Cross-Modality Subregion Transfer
Sep 11, 2026Ziliang Hong, Hongyi Pan, Halil Ertugrul Aktas +7Domain-Adversarial Learning Cross-Modality Medical Image Segmentation
Department of Radiology, Northwestern University, Chicago, IL, USA · Division of Gastroenterology and Hepatology, Mayo Clinic Florida, Jacksonville, FL, USA · Department of Gastroenterology and Hepatology, Northwestern University, Chicago, IL, USA

Oct 5, 2026J/K move · Enter open · S save

Topology-Informed Prompt-Conditioned Universal Segmentation of Uterine Structures from Ultrasound and MRI

Yongheng Sun, Yuexi Gu, Jingwen Sun, Maureen Kohi+1

Department of Radiology and BRIC, UNC at Chapel Hill, Chapel Hill, NC 27599, USA

Ultrasound Image Segmentation Cross-Modality Medical Image Segmentation Promptable Image Segmentation Topology-Preserving Image Segmentation Medical Image Segmentation

Open paper AskPDF arXiv

Multi-structure segmentation of the uterus is important for computer-assisted screening, diagnosis, and treatment planning of uterine diseases, where ultrasound and MRI provide complementary clinical information. However, developing a unified model across these modalities is challenging due to their substantially different image appearances, anatomical contexts, spatial resolutions, and label spaces. Moreover, existing datasets often define different segmentation targets, making joint learning challenging and potentially leading to negative transfer across heterogeneous tasks. To this end, we propose a Topology-informed Prompt-conditioned Universal Segmentation (TPUS) framework for segmenting multiple uterine structures across ultrasound and MRI. TPUS introduces a graph-based multi-dataset backbone comprising modality-specific stems and a modality-shared graph-based encoder-decoder to support modality-sensitive input adaptation, structural feature reasoning, and joint representation learning across heterogeneous uterine segmentation tasks. In addition, TPUS uses task-aware class prompts to condition the segmentation process for different datasets and label spaces, a dynamic convolutional adaptation module to generate task-specific output responses, and a topology-informed loss to encourage anatomically consistent predictions. Experiments on a uterine ultrasound dataset and a T2-weighted uterine myoma MRI dataset demonstrate that TPUS achieves Dice scores of 0.898 and 0.693 on the two held-out test sets, respectively, outperforming several generic and universal segmentation baselines. Source code can be accessed at https://github.com/YonghengSun1997/TPUS.

Abstract

Cross-Modality Medical Image Segmentation

Explore similar work

Topology-Informed Prompt-Conditioned Universal Segmentation of Uterine Structures from Ultrasound and MRI

OvAi Focus: AI-based Multi-class Segmentation of Functional Ovaries and Adnexal Masses in Gynecological Ultrasound

Unified CT and MRI Pancreas Segmentation for Label-Efficient Cross-Modality Subregion Transfer