VISTA: Variance-Gated Inter-Sequence Test-Time Adaptation for Multi-Sequence MRI Segmentation

Haolin Wang; Jiale Zhou; Wenhan Jiang; Xun Lin; Yafei Ou; Yefeng Zheng; Zhipeng Deng

arxiv: 2605.17433 · v1 · pith:OPQCIHNYnew · submitted 2026-05-17 · 💻 cs.CV

VISTA: Variance-Gated Inter-Sequence Test-Time Adaptation for Multi-Sequence MRI Segmentation

Zhipeng Deng , Jiale Zhou , Wenhan Jiang , Haolin Wang , Xun Lin , Yafei Ou , Yefeng Zheng This is my paper

Pith reviewed 2026-05-20 13:06 UTC · model grok-4.3

classification 💻 cs.CV

keywords test-time adaptationmulti-sequence MRIsegmentationinter-sequence consistencypseudo-labelingdomain shiftmedical image analysis

0 comments

The pith

Test-time adaptation for multi-sequence MRI segmentation improves by generating consistency probes through cross-sequence spectrum and patch swaps while gating pseudo-labels with cross-view disagreement variance.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper aims to show that standard test-time adaptation fails for multi-sequence MRI because it ignores shifts in how sequences interact with one another rather than just changes within each sequence. It introduces a generator that creates probe images by exchanging low-frequency spectral content and entropy-focused patches between sequences, which keeps core anatomy the same but stresses the model's inter-sequence reasoning. A separate module then tracks how much different views of the same scan disagree and uses that variance to decide which parts of the model's own predictions are trustworthy enough to retrain on. If the approach holds, segmentation models trained on one set of adult brain scans could deliver better results on scans from children or from low-field scanners in new regions without ever revisiting the original training data.

Core claim

The central claim is that modality-interaction shifts can be handled in a source-free setting by an Inter-Sequence Intervention Generator that produces consistency probes via low-frequency spectrum swapping and entropy-localized patch swapping, combined with Cross-View Disagreement-Aware Pseudo Labeling that computes voxel-wise reliability from cross-view variance to dynamically gate self-training and enforce interventional consistency on robust anatomical semantics.

What carries the argument

The Inter-Sequence Intervention Generator that creates consistency probes by swapping low-frequency spectra and entropy-localized patches across sequences, together with variance-based gating of pseudo-labels drawn from cross-view disagreement.

If this is right

The adapted model records absolute Dice gains of 1.89 percent on low-field African data and 2.82 percent on pediatric data relative to the source model.
Inter-sequence consistency is enforced during adaptation rather than treated as a per-sequence shift.
Voxel-wise pseudo-label selection becomes dynamic through disagreement variance, limiting the use of unreliable predictions for self-training.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same probe-generation idea could be tested on other paired imaging modalities such as CT and PET to see whether cross-modality consistency improves adaptation there as well.
Variance gating of pseudo-labels might stabilize self-training in non-medical settings where multiple sensor streams must stay consistent under domain change.
If the probes reliably isolate inter-sequence effects, the method points toward lighter-weight test-time updates that avoid full retraining for each new scanner or patient population.

Load-bearing premise

The design assumes that swapping low-frequency spectra and entropy-localized patches across sequences preserves anatomical semantics while still breaking unwanted inter-sequence dependencies.

What would settle it

Run the full adaptation pipeline on a held-out multi-sequence MRI dataset with expert ground-truth labels and measure whether the final Dice score on the target cohort is higher than the unadapted source model; no improvement or a drop would show the claimed benefit does not occur.

Figures

Figures reproduced from arXiv: 2605.17433 by Haolin Wang, Jiale Zhou, Wenhan Jiang, Xun Lin, Yafei Ou, Yefeng Zheng, Zhipeng Deng.

**Figure 2.** Figure 2: The pipeline of our VISTA. ISIG simulates modality-interaction shifts via frequency (LFCCS) and spatial (UGPS) cross-sequence swaps. The resulting cross-view disagreement variance dynamically gates the teacher’s anchor pseudo-labels, preventing error accumulation and providing reliable supervision for student optimization. adapting from standard adult MRI (BraTS-GLI-Pre) to African low-field (BraTSSSA) an… view at source ↗

read the original abstract

Deploying multi-sequence magnetic resonance imaging (MRI) segmentation models to new clinical environments is challenging due to variations in scanners and acquisition protocols. Although existing TTA methods handle basic per-modality shifts, they often fail under a fundamental dual-shift problem, as their adaptation signals fail to capture modality-interaction shifts that disrupt inter-sequence consistency. To address this, we propose Variance-gated Inter-Sequence Test-time Adaptation (VISTA), a source-free framework that tackles modality-interaction shifts. First, we design an Inter-Sequence Intervention Generator (ISIG) that generates a set of consistency probes by swapping low-frequency spectra and entropy-localized patches across sequences, preserving anatomical semantics while challenging inter-sequence dependencies. Second, we introduce Cross-View Disagreement-Aware Pseudo Labeling (CDPL), which establishes a voxel-wise reliability metric using cross-view disagreement variance to dynamically gate self-training and enforce interventional consistency, encouraging the network to rely on robust anatomical semantics. Extensive experiments adapting from standard adult MRI (BraTS-GLI-Pre) to African low-field (BraTS-SSA) and pediatric (BraTS-PED) cohorts show improved performance over competing methods under clinical shifts, achieving absolute Dice improvements of +1.89% (SSA) and +2.82% (PED) over the source model. The code is available at https://github.com/dzp2095/VISTA.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

VISTA pairs spectrum and patch swaps with variance-gated pseudo-labeling to target inter-sequence shifts in MRI TTA, showing modest Dice gains on two external cohorts but leaving the key semantic-preservation assumption lightly checked.

read the letter

VISTA's core idea is to handle the dual-shift problem in multi-sequence MRI by creating consistency probes during test time. It swaps low-frequency spectra and entropy-localized patches across sequences like T1 and T2, then uses cross-view disagreement variance to gate which pseudo-labels get used for self-training. This is meant to push the model toward robust anatomical features instead of brittle inter-sequence correlations that break under new scanners or patient groups.

Referee Report

1 major / 1 minor

Summary. The manuscript presents VISTA, a source-free test-time adaptation framework for multi-sequence MRI segmentation addressing modality-interaction shifts. It introduces the Inter-Sequence Intervention Generator (ISIG) to generate consistency probes via low-frequency spectrum swaps and entropy-localized patch exchanges across sequences, and Cross-View Disagreement-Aware Pseudo Labeling (CDPL) that uses voxel-wise cross-view disagreement variance to dynamically gate self-training and enforce interventional consistency. Experiments adapting a source model trained on BraTS-GLI-Pre to external African low-field (BraTS-SSA) and pediatric (BraTS-PED) cohorts report absolute Dice gains of +1.89% and +2.82% over the source model, with code released at https://github.com/dzp2095/VISTA.

Significance. If the central claims hold, the work offers a targeted approach to inter-sequence consistency in test-time adaptation for multi-modal MRI, which is relevant for clinical deployment under scanner and protocol variations. The source-free design and variance-based gating mechanism are practical strengths. Reproducibility is supported by the public code release. The empirical focus on challenging external cohorts (low-field and pediatric) adds value, though the contribution is primarily algorithmic and empirical rather than providing parameter-free derivations or machine-checked proofs.

major comments (1)

[Method overview (ISIG)] Method overview (ISIG description): The core assumption that low-frequency spectrum swaps and entropy-localized patch exchanges preserve anatomical semantics while sufficiently challenging inter-sequence dependencies is load-bearing for the claim that CDPL enforces robust consistency rather than propagating artifacts. No quantitative verification is provided, such as prediction agreement rates or perceptual distance metrics on the intervened images. Without this, the reported Dice improvements (+1.89% on SSA, +2.82% on PED) could arise from incidental regularization effects instead of the intended handling of modality-interaction shifts.

minor comments (1)

[Abstract] Abstract: The description of competing methods and experimental controls could be expanded slightly for better context on the reported gains.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address the major comment below and have revised the paper to incorporate additional verification as suggested.

read point-by-point responses

Referee: The core assumption that low-frequency spectrum swaps and entropy-localized patch exchanges preserve anatomical semantics while sufficiently challenging inter-sequence dependencies is load-bearing for the claim that CDPL enforces robust consistency rather than propagating artifacts. No quantitative verification is provided, such as prediction agreement rates or perceptual distance metrics on the intervened images. Without this, the reported Dice improvements (+1.89% on SSA, +2.82% on PED) could arise from incidental regularization effects instead of the intended handling of modality-interaction shifts.

Authors: We thank the referee for highlighting this important point. We agree that direct quantitative verification of semantic preservation under the ISIG interventions would strengthen the manuscript and help distinguish the intended effect from incidental regularization. In the revised version, we have added experiments that measure prediction agreement rates (using the frozen source model) between original and intervened images, as well as perceptual metrics including SSIM and LPIPS on the generated consistency probes. These results are now reported in Section 4.2 and the supplementary material. The added analyses show that the interventions maintain high semantic consistency while introducing meaningful inter-sequence challenges. We further note that VISTA's gains exceed those of competing TTA methods employing alternative regularization strategies, supporting that the improvements arise from targeted handling of modality-interaction shifts rather than generic effects. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical method validated on held-out external datasets

full rationale

The paper introduces an algorithmic TTA framework (ISIG for generating consistency probes via spectrum/patch swaps, CDPL for variance-gated pseudo-labeling) and measures absolute Dice gains on independent target cohorts (BraTS-SSA, BraTS-PED) distinct from the source training data. No equations reduce a claimed prediction to a fitted parameter by construction, no load-bearing self-citation chains appear, and the central claims rest on external benchmark performance rather than internal redefinitions or ansatzes smuggled via prior author work. This is the standard case of a self-contained empirical contribution.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The framework rests on MRI-domain assumptions about frequency content and prediction reliability rather than new physical entities or free parameters explicitly fitted in the abstract.

free parameters (1)

intervention and gating hyperparameters
Parameters controlling spectrum swap strength and variance threshold in ISIG and CDPL are expected but not quantified in the abstract.

axioms (2)

domain assumption Swapping low-frequency spectra and entropy-localized patches preserves anatomical semantics while disrupting inter-sequence dependencies
Invoked in the description of the Inter-Sequence Intervention Generator.
domain assumption Voxel-wise cross-view disagreement variance reliably indicates prediction reliability for pseudo-label gating
Basis for the Cross-View Disagreement-Aware Pseudo Labeling module.

pith-pipeline@v0.9.0 · 5802 in / 1425 out tokens · 53590 ms · 2026-05-20T13:06:47.334773+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

ISIG generates consistency probes by swapping low-frequency spectra and entropy-localized patches across sequences, preserving anatomical semantics while challenging inter-sequence dependencies.
IndisputableMonolith/Foundation/DimensionForcing.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Cross-View Disagreement Variance Vc(v) = Var_k(p(k)_c(v)) gates pseudo-labeling.

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

30 extracted references · 30 canonical work pages · 3 internal anchors

[1]

Radiology: Artificial Intelligence7(4), e240528 (2025)

Adewole, M., Rudie, J.D., Gbadamosi, A., Zhang, D., Raymond, C., Ajigboto- shso, J., Toyobo, O., Aguh, K., Omidiji, O., Akinola, R., et al.: The BraTS-africa dataset: expanding the brain tumor segmentation data to capture african popula- tions. Radiology: Artificial Intelligence7(4), e240528 (2025)

work page 2025
[2]

Advancing The Cancer Genome Atlas glioma MRI collections with expert segmentation labels and radiomic features,

Bakas, S., Akbari, H., Sotiras, A., Bilello, M., Rozycki, M., Kirby, J., et al.: Advancing the cancer genome atlas glioma MRI collections with expert seg- mentation labels and radiomic features. Scientific Data4, 170117 (2017). https://doi.org/10.1038/sdata.2017.117

work page doi:10.1038/sdata.2017.117 2017
[3]

In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)

Cao, Q., Zeng, H., Gao, P., Wu, R., Zhang, D., Zhao, B., Li, Y., Niu, H., Wang, X., Li, Z.: Multi-modal continual test-time adaptation for 3d semantic segmentation. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). pp. 18809–18819 (October 2023)

work page 2023
[4]

MONAI: An open-source framework for deep learning in healthcare

Cardoso, M.J., Li, W., Brown, R., Ma, N., Kerfoot, E., Wang, Y., Murrey, B., Myronenko, A., Zhao, C., Yang, D., et al.: Monai: An open-source framework for deep learning in healthcare. arXiv preprint arXiv:2211.02701 (2022)

work page internal anchor Pith review Pith/arXiv arXiv 2022
[5]

In: Proceedings of the 42nd International Conference on Machine Learning (ICML)

Chen, C., Huang, Y., Du, Y., Chen, B., Fu, Z., Ghanem, B.: Test-time selective adaptation for uni-modal distribution shift in multi-modal data. In: Proceedings of the 42nd International Conference on Machine Learning (ICML). Proceedings of Machine Learning Research, vol. 267, pp. 9894–9922. PMLR (2025),https: //proceedings.mlr.press/v267/chen25ch.html 10 ...

work page 2025
[6]

In: ProceedingsoftheIEEE/CVFConferenceonComputerVisionandPatternRecog- nition (CVPR)

Chen, Z., Pan, Y., Ye, Y., Lu, M., Xia, Y.: Each test image deserves a specific prompt: Continual test-time adaptation for 2d medical image segmentation. In: ProceedingsoftheIEEE/CVFConferenceonComputerVisionandPatternRecog- nition (CVPR). pp. 11184–11193 (2024)

work page 2024
[7]

, author Abdulkadir, A

Çiçek, Ö., Abdulkadir, A., Lienkamp, S.S., Brox, T., Ronneberger, O.: 3d u- net: Learning dense volumetric segmentation from sparse annotation. In: Medi- cal Image Computing and Computer-Assisted Intervention – MICCAI 2016. Lec- ture Notes in Computer Science, vol. 9901, pp. 424–432. Springer, Cham (2016). https://doi.org/10.1007/978-3-319-46723-8_49

work page doi:10.1007/978-3-319-46723-8_49 2016
[8]

Medical Image Analysis p

Deng, Z., Xu, Z., Isshiki, T., Zheng, Y.: Fedsemidg: Domain generalized federated semi-supervised medical image segmentation. Medical Image Analysis p. 104096 (2026)

work page 2026
[9]

In: Medical Image Computing and Computer Assisted Intervention – MICCAI 2025

Joshi, S., Osuala, R., Garrucho, L., Kushibar, K., Kessler, D., Diaz, O., Lekadir, K.: MuVi: Single image test-time adaptation via multi-view co-training. In: Medical Image Computing and Computer Assisted Intervention – MICCAI 2025. Springer (2025). https://doi.org/10.1007/978-3-032-04978-0_59

work page doi:10.1007/978-3-032-04978-0_59 2025
[10]

In: European Conference on Computer Vision (ECCV) (2024)

Kang, J., Kim, N., Ok, J., Kwak, S.: MemBN: Robust test-time adaptation via batch norm with statistics memory. In: European Conference on Computer Vision (ECCV) (2024)

work page 2024
[11]

Medical Image Analysis68, 101907 (2021)

Karani, N., Erdil, E., Chaitanya, K., Konukoglu, E.: Test-time adaptable neu- ral networks for robust medical image segmentation. Medical Image Analysis68, 101907 (2021)

work page 2021
[12]

https://doi.org/10.48550/arXiv.2407.08855, accepted in Machine Learning for Biomedical Imaging (MELBA), 2025

Kazerooni, A.F., Khalili, N., Liu, X., Haldar, D., Jiang, Z., et al.: BraTS-PEDs: Results of the multi-consortium international pediatric brain tumor segmenta- tion challenge 2023 (2024). https://doi.org/10.48550/arXiv.2407.08855, accepted in Machine Learning for Biomedical Imaging (MELBA), 2025

work page doi:10.48550/arxiv.2407.08855 2023
[13]

Simple and Scalable Predictive Uncertainty Estimation using Deep Ensembles

Lakshminarayanan, B., Pritzel, A., Blundell, C.: Simple and scalable predictive uncertainty estimation using deep ensembles. In: Advances in Neural Information Processing Systems. vol. 30 (2017),https://arxiv.org/abs/1612.01474

work page internal anchor Pith review Pith/arXiv arXiv 2017
[14]

The Multimodal Brain Tumor Image Segmentation Benchmark (BRATS),

Menze, B.H., Jakab, A., Bauer, S., Kalpathy-Cramer, J., Farahani, K., Kirby, J., et al.: The multimodal brain tumor image segmentation benchmark (BraTS). IEEE Transactions on Medical Imaging34(10), 1993–2024 (2015). https://doi.org/10.1109/TMI.2014.2377694

work page doi:10.1109/tmi.2014.2377694 1993
[15]

Niu, S., Wu, J., Zhang, Y., Chen, Y., Zheng, S., Zhao, P., Tan, M.: Efficient test- timemodeladaptationwithoutforgetting.In:InternationalConferenceonMachine Learning (ICML) (2022)

work page 2022
[16]

In: Advances in Neural Information Processing Systems

Sohn, K., Berthelot, D., Li, C.L., Zhang, Z., Carlini, N., Cubuk, E.D., Kurakin, A., Rafailov, R.: FixMatch: Simplifying semi-supervised learning with consistency and confidence. In: Advances in Neural Information Processing Systems. vol. 33 (2020),https://arxiv.org/abs/2001.07685

work page arXiv 2020
[17]

Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results

Tarvainen, A., Valpola, H.: Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results. In: Advances in Neural Information Processing Systems. vol. 30 (2017),https://arxiv.org/abs/ 1703.01780

work page internal anchor Pith review Pith/arXiv arXiv 2017
[18]

Wang, D., Shelhamer, E., Liu, S., Olshausen, B., Darrell, T.: Tent: Fully test- timeadaptationbyentropyminimization.In:InternationalConferenceonLearning Representations (ICLR) (2021)

work page 2021
[19]

In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

Wang, Q., Fink, O., Van Gool, L., Dai, D.: Continual test-time domain adaptation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 7201–7211 (2022) VISTA 11

work page 2022
[20]

arXiv preprint arXiv:2509.17925 (2025)

Wang,Y.,Chen,Y.,Jiang,S., Yu,W.,Liu,M., Wu,B.,Zong,J.,Qin,F., Wang,C., Tian, Q.: SmaRT: Style-modulated robust test-time adaptation for cross-domain brain tumor segmentation in mri. arXiv preprint arXiv:2509.17925 (2025)

work page arXiv 2025
[21]

In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2020)

Yang, Y., Soatto, S.: Fda: Fourier domain adaptation for semantic segmentation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2020)

work page 2020
[22]

arXiv preprint arXiv:2603.03956 (2026)

You, J., Cheng, J., Zhang, J., Zhou, Y.: Towards generalized multimodal homog- raphy estimation. arXiv preprint arXiv:2603.03956 (2026)

work page arXiv 2026
[23]

arXiv preprint arXiv:2512.02497 (2025)

Yu, W., Jiang, S., Chen, Y., Chang, S., Wang, Y., Wu, B., Dong, J., Liu, M., Zhu, S., Qin, F., Wang, C., Tian, Q.: A large scale benchmark for test time adaptation methods in medical image segmentation. arXiv preprint arXiv:2512.02497 (2025)

work page arXiv 2025
[24]

, author Han, D

Yun, S., Han, D., Oh, S.J., Chun, S., Choe, J., Yoo, Y.: CutMix: Regularization strategy to train strong classifiers with localizable features. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). pp. 6023–6032 (2019). https://doi.org/10.1109/ICCV.2019.00612

work page doi:10.1109/iccv.2019.00612 2019
[25]

In: Domain Adaptation and Rep- resentation Transfer (DART 2022), Held in Conjunction with MICCAI 2022

Zakazov, I., Shaposhnikov, V., Bespalov, I., Dylov, D.V.: Feather-light fourier do- main adaptation in magnetic resonance imaging. In: Domain Adaptation and Rep- resentation Transfer (DART 2022), Held in Conjunction with MICCAI 2022. Lec- ture Notes in Computer Science, vol. 13542, pp. 88–97. Springer, Cham (2022). https://doi.org/10.1007/978-3-031-16852-9_9

work page doi:10.1007/978-3-031-16852-9_9 2022
[26]

PLOS Medicine15(11), e1002683 (2018)

Zech, J.R., Badgeley, M.A., Liu, M., Costa, A.B., Titano, J.J., Oermann, E.K.: Variable generalization performance of a deep learning model to detect pneumonia in chest radiographs: A cross-sectional study. PLOS Medicine15(11), e1002683 (2018). https://doi.org/10.1371/journal.pmed.1002683

work page doi:10.1371/journal.pmed.1002683 2018
[27]

IEEE Transactions on Medical Imaging44(4), 1853–1865 (2025)

Zhang, C., Zheng, H., You, X., Zheng, Y., Gu, Y.: PASS: Test-time prompting to adapt styles and semantic shapes in medical image segmen- tation. IEEE Transactions on Medical Imaging44(4), 1853–1865 (2025). https://doi.org/10.1109/TMI.2024.3521463

work page doi:10.1109/tmi.2024.3521463 2025
[28]

In: International Conference on Learning Representations (ICLR) (2025)

Zhang, Q., Bian, Y., Kong, X., Zhao, P., Zhang, C.: COME: Test-time adaption by conservatively minimizing entropy. In: International Conference on Learning Representations (ICLR) (2025)

work page 2025
[29]

Medical Image Analysis92, 103069 (2024)

Zhang, Y., Zhou, T., Tao, Y., Wang, S., Wu, Y., Liu, B., Gu, P., Chen, Q., Chen, D.Z.: Testfit: A plug-and-play one-pass test time method for medical image seg- mentation. Medical Image Analysis92, 103069 (2024)

work page 2024
[30]

In: Proceedings of the IEEE/CVF International Conference on Computer Vision

Zhou, J., Wang, W., Li, S., Qu, X., Guo, X., Liu, Y., Tang, W., Lin, X., Zheng, Y.: Topotta: Topology-enhanced test-time adaptation for tubular structure segmen- tation. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 24123–24134 (2025)

work page 2025

[1] [1]

Radiology: Artificial Intelligence7(4), e240528 (2025)

Adewole, M., Rudie, J.D., Gbadamosi, A., Zhang, D., Raymond, C., Ajigboto- shso, J., Toyobo, O., Aguh, K., Omidiji, O., Akinola, R., et al.: The BraTS-africa dataset: expanding the brain tumor segmentation data to capture african popula- tions. Radiology: Artificial Intelligence7(4), e240528 (2025)

work page 2025

[2] [2]

Advancing The Cancer Genome Atlas glioma MRI collections with expert segmentation labels and radiomic features,

Bakas, S., Akbari, H., Sotiras, A., Bilello, M., Rozycki, M., Kirby, J., et al.: Advancing the cancer genome atlas glioma MRI collections with expert seg- mentation labels and radiomic features. Scientific Data4, 170117 (2017). https://doi.org/10.1038/sdata.2017.117

work page doi:10.1038/sdata.2017.117 2017

[3] [3]

In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)

Cao, Q., Zeng, H., Gao, P., Wu, R., Zhang, D., Zhao, B., Li, Y., Niu, H., Wang, X., Li, Z.: Multi-modal continual test-time adaptation for 3d semantic segmentation. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). pp. 18809–18819 (October 2023)

work page 2023

[4] [4]

MONAI: An open-source framework for deep learning in healthcare

Cardoso, M.J., Li, W., Brown, R., Ma, N., Kerfoot, E., Wang, Y., Murrey, B., Myronenko, A., Zhao, C., Yang, D., et al.: Monai: An open-source framework for deep learning in healthcare. arXiv preprint arXiv:2211.02701 (2022)

work page internal anchor Pith review Pith/arXiv arXiv 2022

[5] [5]

In: Proceedings of the 42nd International Conference on Machine Learning (ICML)

Chen, C., Huang, Y., Du, Y., Chen, B., Fu, Z., Ghanem, B.: Test-time selective adaptation for uni-modal distribution shift in multi-modal data. In: Proceedings of the 42nd International Conference on Machine Learning (ICML). Proceedings of Machine Learning Research, vol. 267, pp. 9894–9922. PMLR (2025),https: //proceedings.mlr.press/v267/chen25ch.html 10 ...

work page 2025

[6] [6]

In: ProceedingsoftheIEEE/CVFConferenceonComputerVisionandPatternRecog- nition (CVPR)

Chen, Z., Pan, Y., Ye, Y., Lu, M., Xia, Y.: Each test image deserves a specific prompt: Continual test-time adaptation for 2d medical image segmentation. In: ProceedingsoftheIEEE/CVFConferenceonComputerVisionandPatternRecog- nition (CVPR). pp. 11184–11193 (2024)

work page 2024

[7] [7]

, author Abdulkadir, A

Çiçek, Ö., Abdulkadir, A., Lienkamp, S.S., Brox, T., Ronneberger, O.: 3d u- net: Learning dense volumetric segmentation from sparse annotation. In: Medi- cal Image Computing and Computer-Assisted Intervention – MICCAI 2016. Lec- ture Notes in Computer Science, vol. 9901, pp. 424–432. Springer, Cham (2016). https://doi.org/10.1007/978-3-319-46723-8_49

work page doi:10.1007/978-3-319-46723-8_49 2016

[8] [8]

Medical Image Analysis p

Deng, Z., Xu, Z., Isshiki, T., Zheng, Y.: Fedsemidg: Domain generalized federated semi-supervised medical image segmentation. Medical Image Analysis p. 104096 (2026)

work page 2026

[9] [9]

In: Medical Image Computing and Computer Assisted Intervention – MICCAI 2025

Joshi, S., Osuala, R., Garrucho, L., Kushibar, K., Kessler, D., Diaz, O., Lekadir, K.: MuVi: Single image test-time adaptation via multi-view co-training. In: Medical Image Computing and Computer Assisted Intervention – MICCAI 2025. Springer (2025). https://doi.org/10.1007/978-3-032-04978-0_59

work page doi:10.1007/978-3-032-04978-0_59 2025

[10] [10]

In: European Conference on Computer Vision (ECCV) (2024)

Kang, J., Kim, N., Ok, J., Kwak, S.: MemBN: Robust test-time adaptation via batch norm with statistics memory. In: European Conference on Computer Vision (ECCV) (2024)

work page 2024

[11] [11]

Medical Image Analysis68, 101907 (2021)

Karani, N., Erdil, E., Chaitanya, K., Konukoglu, E.: Test-time adaptable neu- ral networks for robust medical image segmentation. Medical Image Analysis68, 101907 (2021)

work page 2021

[12] [12]

https://doi.org/10.48550/arXiv.2407.08855, accepted in Machine Learning for Biomedical Imaging (MELBA), 2025

Kazerooni, A.F., Khalili, N., Liu, X., Haldar, D., Jiang, Z., et al.: BraTS-PEDs: Results of the multi-consortium international pediatric brain tumor segmenta- tion challenge 2023 (2024). https://doi.org/10.48550/arXiv.2407.08855, accepted in Machine Learning for Biomedical Imaging (MELBA), 2025

work page doi:10.48550/arxiv.2407.08855 2023

[13] [13]

Simple and Scalable Predictive Uncertainty Estimation using Deep Ensembles

Lakshminarayanan, B., Pritzel, A., Blundell, C.: Simple and scalable predictive uncertainty estimation using deep ensembles. In: Advances in Neural Information Processing Systems. vol. 30 (2017),https://arxiv.org/abs/1612.01474

work page internal anchor Pith review Pith/arXiv arXiv 2017

[14] [14]

The Multimodal Brain Tumor Image Segmentation Benchmark (BRATS),

Menze, B.H., Jakab, A., Bauer, S., Kalpathy-Cramer, J., Farahani, K., Kirby, J., et al.: The multimodal brain tumor image segmentation benchmark (BraTS). IEEE Transactions on Medical Imaging34(10), 1993–2024 (2015). https://doi.org/10.1109/TMI.2014.2377694

work page doi:10.1109/tmi.2014.2377694 1993

[15] [15]

Niu, S., Wu, J., Zhang, Y., Chen, Y., Zheng, S., Zhao, P., Tan, M.: Efficient test- timemodeladaptationwithoutforgetting.In:InternationalConferenceonMachine Learning (ICML) (2022)

work page 2022

[16] [16]

In: Advances in Neural Information Processing Systems

Sohn, K., Berthelot, D., Li, C.L., Zhang, Z., Carlini, N., Cubuk, E.D., Kurakin, A., Rafailov, R.: FixMatch: Simplifying semi-supervised learning with consistency and confidence. In: Advances in Neural Information Processing Systems. vol. 33 (2020),https://arxiv.org/abs/2001.07685

work page arXiv 2020

[17] [17]

Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results

Tarvainen, A., Valpola, H.: Mean teachers are better role models: Weight-averaged consistency targets improve semi-supervised deep learning results. In: Advances in Neural Information Processing Systems. vol. 30 (2017),https://arxiv.org/abs/ 1703.01780

work page internal anchor Pith review Pith/arXiv arXiv 2017

[18] [18]

Wang, D., Shelhamer, E., Liu, S., Olshausen, B., Darrell, T.: Tent: Fully test- timeadaptationbyentropyminimization.In:InternationalConferenceonLearning Representations (ICLR) (2021)

work page 2021

[19] [19]

In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

Wang, Q., Fink, O., Van Gool, L., Dai, D.: Continual test-time domain adaptation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 7201–7211 (2022) VISTA 11

work page 2022

[20] [20]

arXiv preprint arXiv:2509.17925 (2025)

Wang,Y.,Chen,Y.,Jiang,S., Yu,W.,Liu,M., Wu,B.,Zong,J.,Qin,F., Wang,C., Tian, Q.: SmaRT: Style-modulated robust test-time adaptation for cross-domain brain tumor segmentation in mri. arXiv preprint arXiv:2509.17925 (2025)

work page arXiv 2025

[21] [21]

In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2020)

Yang, Y., Soatto, S.: Fda: Fourier domain adaptation for semantic segmentation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) (2020)

work page 2020

[22] [22]

arXiv preprint arXiv:2603.03956 (2026)

You, J., Cheng, J., Zhang, J., Zhou, Y.: Towards generalized multimodal homog- raphy estimation. arXiv preprint arXiv:2603.03956 (2026)

work page arXiv 2026

[23] [23]

arXiv preprint arXiv:2512.02497 (2025)

Yu, W., Jiang, S., Chen, Y., Chang, S., Wang, Y., Wu, B., Dong, J., Liu, M., Zhu, S., Qin, F., Wang, C., Tian, Q.: A large scale benchmark for test time adaptation methods in medical image segmentation. arXiv preprint arXiv:2512.02497 (2025)

work page arXiv 2025

[24] [24]

, author Han, D

Yun, S., Han, D., Oh, S.J., Chun, S., Choe, J., Yoo, Y.: CutMix: Regularization strategy to train strong classifiers with localizable features. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). pp. 6023–6032 (2019). https://doi.org/10.1109/ICCV.2019.00612

work page doi:10.1109/iccv.2019.00612 2019

[25] [25]

In: Domain Adaptation and Rep- resentation Transfer (DART 2022), Held in Conjunction with MICCAI 2022

Zakazov, I., Shaposhnikov, V., Bespalov, I., Dylov, D.V.: Feather-light fourier do- main adaptation in magnetic resonance imaging. In: Domain Adaptation and Rep- resentation Transfer (DART 2022), Held in Conjunction with MICCAI 2022. Lec- ture Notes in Computer Science, vol. 13542, pp. 88–97. Springer, Cham (2022). https://doi.org/10.1007/978-3-031-16852-9_9

work page doi:10.1007/978-3-031-16852-9_9 2022

[26] [26]

PLOS Medicine15(11), e1002683 (2018)

Zech, J.R., Badgeley, M.A., Liu, M., Costa, A.B., Titano, J.J., Oermann, E.K.: Variable generalization performance of a deep learning model to detect pneumonia in chest radiographs: A cross-sectional study. PLOS Medicine15(11), e1002683 (2018). https://doi.org/10.1371/journal.pmed.1002683

work page doi:10.1371/journal.pmed.1002683 2018

[27] [27]

IEEE Transactions on Medical Imaging44(4), 1853–1865 (2025)

Zhang, C., Zheng, H., You, X., Zheng, Y., Gu, Y.: PASS: Test-time prompting to adapt styles and semantic shapes in medical image segmen- tation. IEEE Transactions on Medical Imaging44(4), 1853–1865 (2025). https://doi.org/10.1109/TMI.2024.3521463

work page doi:10.1109/tmi.2024.3521463 2025

[28] [28]

In: International Conference on Learning Representations (ICLR) (2025)

Zhang, Q., Bian, Y., Kong, X., Zhao, P., Zhang, C.: COME: Test-time adaption by conservatively minimizing entropy. In: International Conference on Learning Representations (ICLR) (2025)

work page 2025

[29] [29]

Medical Image Analysis92, 103069 (2024)

Zhang, Y., Zhou, T., Tao, Y., Wang, S., Wu, Y., Liu, B., Gu, P., Chen, Q., Chen, D.Z.: Testfit: A plug-and-play one-pass test time method for medical image seg- mentation. Medical Image Analysis92, 103069 (2024)

work page 2024

[30] [30]

In: Proceedings of the IEEE/CVF International Conference on Computer Vision

Zhou, J., Wang, W., Li, S., Qu, X., Guo, X., Liu, Y., Tang, W., Lin, X., Zheng, Y.: Topotta: Topology-enhanced test-time adaptation for tubular structure segmen- tation. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 24123–24134 (2025)

work page 2025