MICCAI 2026 Satellite Events - Open Access

Physics-informed Unsupervised Test-time Adaptation for MRI Super-Resolution

2026-09-21T00:00:00-04:00

Abstract

Deep learning-based super-resolution (SR) usually does not perform well across domains, e.g., different magnetic resonance imaging (MRI) datasets, usually due to two reasons: 1) there is often a mismatch in terms of data distributions between the training data and test data, and 2) the overly simplified image degradation models in popular SR techniques do not model the underlying physical processes well. Hence, SR models often require extensive supervised fine-tuning when transferring to new datasets (target domains). In this paper we propose UTAR, a physics-informed Unsupervised Test-time domain Adaptation framework for deep learning-based super-resolution for magnetic Resonance imaging. UTAR adapts a pre-trained source-domain SR model to a new target domain (e.g., images from a new hospital) using low-resolution (LR) target-domain images only. This is possible through a new physicsinspired imaging chain degradation model (PIC), which incorporates both spatial and frequency domain (k-space) degradation. Such dualdomain degradation modeling enables us to better capture the variations of LR images that could potentially affect a model’s performance across different unseen MRI domains. We further propose a conditional consistency learning (CCL) strategy to encourage prediction consistency, serving as the supervision signal for TTA. Our method is generic and can accommodate arbitrary pre-trained SR networks. Experimental results verify the effectiveness of UTAR in reducing the performance gap with minimum efforts. UTAR outperforms earlier methods based on different degradation-based strategies and existing test time adaptation (TTA) approaches. We expect our method to provide a key step towards the deployment of MRI SR algorithms in clinical applications where significant domain shifts are inevitable.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ADSMI2024_001.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to Open Review

Open Review Page: Not Available

BibTex

@InProceedings{ZhaWei_Physicsinformed_MICCAISAT2026,
        author = { Zhang, Weitong AND Stelter, Jonathan AND Li, Liu AND Kori, Avinash AND Ouyang, Cheng AND Chen, Chen AND Knoll, Florian AND Karampinos, Dimitrios AND Kainz, Bernhard},
        title = { { Physics-informed Unsupervised Test-time Adaptation for MRI Super-Resolution } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 17276},
        month = {pending},
        page = {pending}
}

back to top

Variational multimodal distillation for diagnosing plaque vulnerability in carotid 3D MRI

2026-09-21T00:00:00-04:00

Abstract

Multimodal learning has attracted much attention in recent years due to its ability to effectively utilize data features from a variety of different modalities. Diagnosing the vulnerability of atherosclerotic plaques directly from carotid 3D MRI images is relatively challenging for both radiologists and conventional 3D vision networks. In clinical prac- tice, radiologists assess patient conditions using a multimodal approach that incorporates various imaging modalities and domain-specific exper- tise, paving the way for the creation of multimodal diagnostic networks. In this paper, we have developed an effective strategy to leverage radi- ologists’ domain knowledge to automate the diagnosis of carotid plaque vulnerability through Variation inference and Multimodal knowledge Distillation (VMD). This method excels in harnessing cross-modality prior knowledge from limited image annotations and radiology reports within training data, thereby enhancing the diagnostic network’s accu- racy for unannotated 3D MRI images. We conducted in-depth experi- ments on the dataset collected in-house and verified the effectiveness of the VMD strategy we proposed. Code will be available at this url.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ADSMI2024_012.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to Open Review

Open Review Page: Not Available

BibTex

@InProceedings{CaoBo_Variational_MICCAISAT2026,
        author = { Cao, Bo AND Feng, Mengmeng AND Yu, Fan AND Qian, Zhen AND Lu, Jie},
        title = { { Variational multimodal distillation for diagnosing plaque vulnerability in carotid 3D MRI } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 17276},
        month = {pending},
        page = {pending}
}

back to top

MedMNIST-C: Comprehensive benchmark and improved classifier robustness by simulating realistic image corruptions

2026-09-21T00:00:00-04:00

Abstract

The integration of neural-network-based systems into clin- ical practice is limited by challenges related to domain generalization androbustness.Thecomputervisioncommunityestablishedbenchmarks such as ImageNet-C as a fundamental prerequisite to measure progress towards those challenges. Similar datasets are largely absent in the med- ical imaging community which lacks a comprehensive benchmark that spans across imaging modalities and applications. To address this gap, we create and open-source MedMNIST-C, a benchmark dataset based on the MedMNIST+ collection, covering 12 datasets and 9 imaging modali- ties. We simulate task and modality-specific image corruptions of varying severity to comprehensively evaluate the robustness of established algo- rithms against real-world artifacts and distribution shifts. We further provide quantitative evidence that our simple-to-use artificial corrup- tions allow for highly performant, lightweight data augmentation to en- hance model robustness. Unlike traditional, generic augmentation strate- gies, our approach leverages domain knowledge, exhibiting significantly higher robustness when compared to widely adopted methods. By intro- ducing MedMNIST-C andopen-sourcingthecorrespondinglibraryallowing for targeted data augmentations, we contribute to the development of in- creasingly robust methods tailored to the challenges of medical imaging. The code is available at github.com/francescodisalvo05/medmnistc-api.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ADSMI2024_022.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to Open Review

Open Review Page: Not Available

BibTex

@InProceedings{DiFra_MedMNISTC_MICCAISAT2026,
        author = { Di Salvo, Francesco AND Doerrich, Sebastian AND Ledig, Christian},
        title = { { MedMNIST-C: Comprehensive benchmark and improved classifier robustness by simulating realistic image corruptions } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 17276},
        month = {pending},
        page = {pending}
}

back to top

CoMoTo: Unpaired Cross-Modal Lesion Distillation Improves Breast Lesion Detection in Tomosynthesis

2026-09-21T00:00:00-04:00

Abstract

Digital Breast Tomosynthesis (DBT) is an advanced breast imagingmodalitythatofferssuperiorlesiondetectionaccuracycompared to conventional mammography, albeit at the trade-off of longer reading time. Accelerating lesion detection from DBT using deep learning is hin- dered by limited data availability and huge annotation costs. A possible solution to this issue could be to leverage the information provided by a more widely available modality, such as mammography, to enhance DBT lesion detection. In this paper, we present a novel framework, Co- MoTo, for improving lesion detection in DBT. Our framework leverages unpaired mammography data to enhance the training of a DBT model, improving practicality by eliminating the need for mammography dur- ing inference. Specifically, we propose two novel components, Lesion- specific Knowledge Distillation (LsKD) and Intra-modal Point Align- ment (ImPA). LsKD selectively distills lesion features from a mammog- raphy teacher model to a DBT student model, disregarding background features. ImPA further enriches LsKD by ensuring the alignment of le- sion features within the teacher before distilling knowledge to the stu- dent. Our comprehensive evaluation shows that CoMoTo is superior to traditional pretraining and image-level KD, improving performance by 7% Mean Sensitivity under low-data setting. Our code is available at https://github.com/Muhammad-Al-Barbary/CoMoTo.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ADSMI2024_033.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to Open Review

Open Review Page: Not Available

BibTex

@InProceedings{AlbMuh_CoMoTo_MICCAISAT2026,
        author = { Alberb, Muhammad AND Elbatel, Marawan AND Elgebaly, Aya AND Montoya-del-Angel, Ricardo AND Li, Xiaomeng AND Martí, Robert},
        title = { { CoMoTo: Unpaired Cross-Modal Lesion Distillation Improves Breast Lesion Detection in Tomosynthesis } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 17276},
        month = {pending},
        page = {pending}
}

back to top

Learning from Similarity Proportion Loss for Classifying Skeletal Muscle Recovery Stages

2026-09-21T00:00:00-04:00

Abstract

Evaluating the regeneration process of damaged muscle tis- sue is a fundamental analysis in muscle research to measure experimen- tal effect sizes and uncover mechanisms behind muscle weakness due to aging and disease. The conventional approach to assessing muscle tissue regeneration involves whole-slide imaging and expert visual inspection of the recovery stages based on the morphological information of cells and fibers. There is a need to replace these tasks with automated methods incorporating machine learning techniques to ensure a quantitative and objective analysis. Given the limited availability of fully labeled data, a possible approach is Learning from Label Proportions (LLP), a weakly supervised learning method using class label proportions. However, cur- rentLLPmethodshavetwolimitations:(1)theycannotadaptthefeature extractor for muscle tissues, and (2) they treat the classes representing recovery stages and cell morphological changes as nominal, resulting in the loss of ordinal information. To address these issues, we propose Or- dinal Scale Learning from Similarity Proportion (OSLSP), which uses a similarity proportion loss derived from two bag combinations. OSLSP can update the feature extractor by using class proportion attention to the ordinal scale of the class. Our model with OSLSP outperforms large- scale pre-trained and fine-tuning models in classification tasks of skeletal muscle recovery stages.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ADSMI2024_044.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/ADSMI2024_044_supp.pdf

Link to Open Review

Open Review Page: Not Available

BibTex

@InProceedings{YamYu_Learning_MICCAISAT2026,
        author = { Yamaoka, Yu AND Chan, Weng Ian AND Seno, Shigeto AND Fukada, Soichiro AND Matsuda, Hideo},
        title = { { Learning from Similarity Proportion Loss for Classifying Skeletal Muscle Recovery Stages } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 17276},
        month = {pending},
        page = {pending}
}

back to top

LoGex: Improved tail detection of extremely rare histopathology classes via guided diffusion

2026-09-21T00:00:00-04:00

Abstract

In realistic medical settings, the data are often inherently long-tailed, with most samples concentrated in a few classes and a long tail of rare classes, usually containing just a few samples. This distribu- tionpresentsasignificantchallengebecauserareconditionsarecriticalto detect and difficult to classify due to limited data. In this paper, rather than attempting to classify rare classes, we aim to detect these as out- of-distribution data reliably. We leverage low-rank adaption (LoRA) and diffusion guidance to generate targeted synthetic data for the detection problem. We significantly improve the OOD detection performance on a challenging histopathological task with only ten samples per tail class without losing classification accuracy on the head classes.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ADSMI2024_054.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/ADSMI2024_054_supp.pdf

Link to Open Review

Open Review Page: Not Available

BibTex

@InProceedings{MülMax_LoGex_MICCAISAT2026,
        author = { Müller, Maximilian AND Hein, Matthias},
        title = { { LoGex: Improved tail detection of extremely rare histopathology classes via guided diffusion } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 17276},
        month = {pending},
        page = {pending}
}

back to top

SAM Carries the Burden: A Semi-Supervised Approach Refining Pseudo Labels for Medical Segmentation

2026-09-21T00:00:00-04:00

Abstract

Semantic segmentation is a crucial task in medical imaging. Although supervised learning techniques have proven to be effective in performing this task, they heavily depend on large amounts of annotated training data. The recently introduced Segment Anything Model (SAM) enables prompt-based segmentation and offers zero-shot generalization to unfamiliar objects. In our work, we leverage SAM’s abstract object understanding for medical image segmentation to provide pseudo labels for semi-supervised learning, thereby mitigating the need for extensive annotated training data. Our approach refines initial segmentations that are derived from a lim- ited amount of annotated data (comprising up to 43 cases) by extracting bounding boxes and seed points as prompts forwarded to SAM. Thus, it enables the generation of dense segmentation masks as pseudo labels for unlabelled data. The results show that training with our pseudo la- bels yields an improvement in Dice score from 74.29 % to 84.17 % and from 66.63 % to 74.87 % for the segmentation of bones of the paedi- atric wrist and teeth in dental radiographs, respectively. As a result, our method outperforms intensity-based post-processing methods, state-of- the-art supervised learning for segmentation (nnU-Net), and the semi- supervised mean teacher approach. Our Code is available on GitHub.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ADSMI2024_065.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/ADSMI2024_065_supp.pdf

Link to Open Review

Open Review Page: Not Available

BibTex

@InProceedings{KeuRon_SAM_MICCAISAT2026,
        author = { Keuth, Ron AND Hansen, Lasse AND Balks, Maren AND Jager, Ronja AND Schroder, Anne-Nele AND Tushaus, Ludger AND Heinrich, Mattias P.},
        title = { { SAM Carries the Burden: A Semi-Supervised Approach Refining Pseudo Labels for Medical Segmentation } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 17276},
        month = {pending},
        page = {pending}
}

back to top

Unsupervised Domain Adaptation for Pediatric Brain Tumor Segmentation

2026-09-21T00:00:00-04:00

Abstract

Significant advances have been made toward building accu- rate automatic segmentation models for adult gliomas. However, the per- formanceofthesemodelsoftendegradeswhenappliedtopediatricglioma due to their imaging and clinical differences (domain shift). Obtaining sufficientannotateddataforpediatricgliomaistypicallydifficultbecause of its rare nature. Also, manual annotations are scarce and expensive. In this work, we propose Domain-Adapted nnU-Net (DA-nnUNet) to perform unsupervised domain adaptation from adult glioma (source do- main) to pediatric glioma (target domain). Specifically, we add a domain classifier connected with a gradient reversal layer (GRL) to a backbone nnU-Net. Once the classifier reaches a very high accuracy, the GRL is activated with the goal of transferring domain-invariant features from the classifier to the segmentation model while preserving segmentation accuracy on the source domain. The accuracy of the classifier slowly degrades to chance levels. No annotations are used in the target do- main. The method is compared to 8 different supervised models using BraTS-Adult glioma (N=1251) and BraTS-PED glioma data (N=99). The proposed method shows notable performance enhancements in the tumor core (TC) region compared to the model that only uses adult data:∼32% better Dice scores and∼20 better 95th percentile Hausdorff distances. Moreover, our unsupervised approach shows no statistically significant difference compared to the practical upper bound model us- ing manual annotations from both datasets in TC region. The code is shared at https://github.com/Fjr9516/DA_nnUNet.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ADSMI2024_076.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to Open Review

Open Review Page: Not Available

BibTex

@InProceedings{FuJin_Unsupervised_MICCAISAT2026,
        author = { Fu, Jingru AND Bendazzoli, Simone AND Smedby, Örjan AND Moreno, Rodrigo},
        title = { { Unsupervised Domain Adaptation for Pediatric Brain Tumor Segmentation } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 17276},
        month = {pending},
        page = {pending}
}

back to top

Rethinking Annotator Simulation: Realistic Evaluation of Whole-Body PET Lesion Interactive Segmentation Methods

2026-09-21T00:00:00-04:00

Abstract

Interactive segmentation plays a crucial role in accelerat- ing the annotation, particularly in domains requiring specialized exper- tise such as nuclear medicine. For example, annotating lesions in whole- body Positron Emission Tomography (PET) images can require over an hour per volume. While previous works evaluate interactive segmenta- tion models through either real user studies or simulated annotators, both approaches present challenges. Real user studies are expensive and often limited in scale, while simulated annotators, also known as robot users, tend to overestimate model performance due to their idealized na- ture. To address these limitations, we introduce four evaluation metrics that quantify the user shift between real and simulated annotators. In an initial user study involving four annotators, we assess existing robot users using our proposed metrics and find that robot users significantly deviate in performance and annotation behavior compared to real anno- tators. Based on these findings, we propose a more realistic robot user that reduces the user shift by incorporating human factors such as click variation and inter-annotator disagreement. We validate our robot user in a second user study, involving four other annotators, and show it con- sistently reduces the simulated-to-real user shift compared to traditional robot users. By employing our robot user, we can conduct more large- scale and cost-efficient evaluations of interactive segmentation models, while preserving the fidelity of real user studies. Our implementation is based on MONAI Label and will be made publicly available.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ADSMI2024_086.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/ADSMI2024_086_supp.pdf

Link to Open Review

Open Review Page: Not Available

BibTex

@InProceedings{MarZdr_Rethinking_MICCAISAT2026,
        author = { Marinov, Zdravko AND Kim, Moon AND Kleesiek, Jens AND Stiefelhagen, Rainer},
        title = { { Rethinking Annotator Simulation: Realistic Evaluation of Whole-Body PET Lesion Interactive Segmentation Methods } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 17276},
        month = {pending},
        page = {pending}
}

back to top

Federated Self-supervised Domain Generalization for Label-efficient Polyp Segmentation

2026-09-21T00:00:00-04:00

Abstract

Employing self-supervised learning (SSL) methodologies as- sumes par-amount significance in handling unlabeled polyp datasets when building deep learning-based automatic polyp segmentation mod- els. However, the intricate privacy dynamics surrounding medical data often preclude seamless data sharing among disparate medical centers. Federated learning (FL) emerges as a formidable solution to this privacy conundrum, yet within the realm of FL, optimizing model generaliza- tion stands as a pressing imperative. Robust generalization capabilities are imperative to ensure the model’s eﬀicacy across diverse geographi- cal domains post-training on localized client datasets. In this paper, a Federated self-supervised Domain Generalization method is proposed to enhance the generalization capacity of federated and Label-eﬀicient intestinal polyp segmentation, named LFDG. Based on a classical SSL method, DropPos, LFDG proposes an adversarial learning-based data augmentation method (SSADA) to enhance the data diversity. LFDG further proposes a relaxation module based on Source-reconstruction and Augmentation-masking (SRAM) to maintain stability in feature learning. We have validated LFDG on polyp images from six medical centers. The performance of our method achieves 3.80% and 3.92% better than the baseline and other recent FL methods and SSL methods, respectively.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ADSMI2024_096.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to Open Review

Open Review Page: Not Available

BibTex

@InProceedings{TanXin_Federated_MICCAISAT2026,
        author = { Tan, Xinyi AND Wang, Jiacheng AND Wang, Liansheng},
        title = { { Federated Self-supervised Domain Generalization for Label-efficient Polyp Segmentation } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 17276},
        month = {pending},
        page = {pending}
}

back to top