arxiv: 2604.17455 · v1 · submitted 2026-04-19 · 💻 cs.CV

Recognition: unknown

From Adaptation to Generalization: Adaptive Visual Prompting for Medical Image Segmentation

Evren \c{C}etinkaya , Sangmin Lee , Jung Uk Kim , Hong Joo Lee , Nassir Navab

Authors on Pith no claims yet

Pith reviewed 2026-05-10 05:24 UTC · model grok-4.3

classification 💻 cs.CV

keywords medical image segmentationvisual promptingdomain generalizationadaptive promptsFourier spectrumcontrastive learningprompt memory

0 comments

The pith

Input-specific prompts retrieved from a memory using Fourier domain features let pre-trained models segment medical images across new domains without retraining.

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

The paper proposes APEX to move beyond single-prompt-per-domain methods by maintaining a memory of diverse prompt representations that can be retrieved on a per-input basis. Domain features pulled from the low-frequency part of each image's Fourier spectrum serve as the query to select the right prompt from that memory. A contrastive training step called Low-Frequency Feature Contrastive learning forces these domain features to cluster by source while remaining separable across sources. Tests on two medical segmentation tasks show the resulting prompts raise accuracy on both familiar and previously unseen domains and add gains to any existing backbone without changing its weights. This matters for medical imaging because images routinely differ within a hospital and between hospitals, so a fixed prompt often fails on new patients or scanners.

Core claim

APEX stores diverse prompt representations in a learnable memory and retrieves an input-specific prompt by matching domain features extracted from the Fourier spectrum of the input image; the memory is trained with Low-Frequency Feature Contrastive learning so that features from the same domain cluster together while features from different domains are pushed apart, yielding better generalization across seen and unseen domains in medical segmentation without updating the underlying model parameters.

What carries the argument

A learnable prompt memory that is queried by domain features taken from the Fourier spectrum of each input image and trained by Low-Frequency Feature Contrastive learning to produce domain-discriminative representations.

If this is right

Segmentation accuracy rises on both domains that were seen during training and on entirely new domains.
The method can be added to any existing segmentation backbone and raises its performance without retraining the backbone weights.
Intra-domain and inter-domain shifts are handled by supplying a different prompt for each input rather than one prompt per entire domain.
No parameter updates to the base model are required, preserving the original model while still adapting to new data distributions.

Where Pith is reading between the lines

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

The same memory-plus-Fourier-query design could be tested on other dense prediction tasks such as medical image registration or lesion detection.
If low-frequency Fourier features prove stable across imaging modalities, the approach might reduce the volume of labeled data needed when a new scanner or patient population appears.
Real-time clinical pipelines could store the prompt memory once and then adapt on the fly to each incoming scan without any model fine-tuning step.

Load-bearing premise

Features extracted from the Fourier spectrum of an image are reliable enough to select the single best prompt from the memory for that image under the variability present in medical data.

What would settle it

A controlled test on a medical segmentation dataset where images from two different domains have nearly identical low-frequency Fourier spectra yet require distinct segmentation boundaries; if performance then drops below a fixed-prompt baseline, the adaptive retrieval mechanism has failed.

Figures

Figures reproduced from arXiv: 2604.17455 by Evren \c{C}etinkaya, Hong Joo Lee, Jung Uk Kim, Nassir Navab, Sangmin Lee.

**Figure 2.** Figure 2: Overview of the proposed adaptive prompting framework. During optimization of APEX, the parameters of the segmentation [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Figure 3: Visual descriptions about the LFC learning framework. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

**Figure 4.** Figure 4: Qualitative comparison between before and after [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗

**Figure 6.** Figure 6: Visualization of the top 10% activated memory slots for [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗

read the original abstract

Visual prompting has emerged as a powerful method for adapting pre-trained models to new domains without updating model parameters. However, existing prompting methods typically optimize a single prompt per domain and apply it uniformly to all inputs, limiting their ability to generalize under intra and inter-domain variability, which is especially critical in the medical field. To address this, we propose APEX, an Adaptive Prompt EXtraction framework that retrieves input-specific prompts from a learnable prompt memory. The memory stores diverse, domain-discriminative prompt representations and is queried via domain features extracted from the Fourier spectrum. To learn robust and discriminative domain features, we introduce a novel Low-Frequency Feature Contrastive (LFC) learning framework that clusters representations from the same domain while separating those from different domains. Extensive experiments on two medical segmentation tasks demonstrate that APEX significantly improves generalization across both seen and unseen domains. Furthermore, it complements any existing backbones and consistently enhances performance, confirming its effectiveness as a plug-and-play prompting solution in medical fields. The code is available at https://github.com/cetinkayaevren/apex/

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

APEX brings input-specific prompt retrieval via Fourier low-frequency features and a new LFC contrastive loss to visual prompting, which is a clear step beyond fixed per-domain prompts, but the generalization gains rest on untested assumptions about what those features actually capture.

read the letter

APEX introduces an adaptive visual prompting approach for medical segmentation that retrieves input-specific prompts from a memory bank using Fourier spectrum features, trained via a low-frequency contrastive loss. The main novelty is combining prompt memory with Fourier-based querying and the LFC framework to handle both intra- and inter-domain shifts without updating the model. This is a practical step beyond fixed per-domain prompts, and the plug-and-play nature plus released code make it easy to test. The paper does a good job highlighting the problem of variability in medical imaging and showing how their method can complement existing backbones. Experiments on two tasks reportedly show gains on seen and unseen domains. That said, the central mechanism depends on low-frequency Fourier features being good enough to pick the right prompts, but these features often pick up on global intensity or acquisition artifacts rather than the subtle anatomical cues that matter for segmentation. If that's the case here, the adaptation might not be as input-specific as claimed. The abstract keeps results at a high level with no mention of statistical tests or detailed ablations, so it's hard to judge how robust the improvements really are. This work is aimed at the medical imaging community working on domain generalization and prompting methods. Anyone building segmentation models for clinical use could find the idea useful to try, especially with the code available. It has enough substance to warrant peer review, though I'd want to see the full experimental details and ablations before citing it myself. I'd recommend sending it for review rather than desk rejecting it.

Referee Report

2 major / 2 minor

Summary. The paper proposes APEX, an Adaptive Prompt EXtraction framework for medical image segmentation that maintains a learnable prompt memory storing domain-discriminative representations. These are queried using domain features extracted from the low-frequency Fourier spectrum of input images, with a novel Low-Frequency Feature Contrastive (LFC) loss introduced to cluster same-domain features and separate different-domain ones. The method is positioned as a plug-and-play adapter that improves generalization to both seen and unseen domains across two medical segmentation tasks without updating backbone parameters, with code released for reproducibility.

Significance. If the central claims hold, the work would provide a meaningful contribution to domain generalization in medical imaging by enabling input-specific visual prompting that addresses intra- and inter-domain variability more flexibly than static per-domain prompts. The plug-and-play nature and public code are strengths that could facilitate adoption and further research in adapting pre-trained models for clinical segmentation tasks.

major comments (2)

[§3.2] §3.2 (domain feature extraction and prompt querying): The central generalization claim depends on low-frequency Fourier features reliably indexing the prompt memory to handle both intra- and inter-domain shifts. No visualization, nearest-neighbor analysis, or failure-case study is provided to demonstrate that these features encode the anatomical or acquisition variations driving segmentation errors rather than global intensity or scanner biases; if the latter, the adaptive mechanism reduces to a domain-level average prompt.
[§4] §4 (experimental validation): The reported performance improvements on unseen domains are described at a high level without statistical significance tests (e.g., paired t-tests or Wilcoxon tests across multiple random seeds), full ablation tables isolating the LFC loss and prompt memory size, or exhaustive baseline comparisons that control for backbone capacity and standard prompting variants. This weakens support for the claim that gains are attributable to the adaptive querying rather than the backbone or generic prompting.

minor comments (2)

[§3.1] Notation for the prompt memory size and LFC weighting coefficients should be introduced explicitly in §3.1 with a table summarizing all free parameters.
[Figure 2] Figure 2 (method overview) would benefit from clearer labeling of the Fourier transform path and the contrastive loss computation to improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment point by point below, providing our responses and indicating the revisions we will incorporate to strengthen the manuscript.

read point-by-point responses

Referee: [§3.2] §3.2 (domain feature extraction and prompt querying): The central generalization claim depends on low-frequency Fourier features reliably indexing the prompt memory to handle both intra- and inter-domain shifts. No visualization, nearest-neighbor analysis, or failure-case study is provided to demonstrate that these features encode the anatomical or acquisition variations driving segmentation errors rather than global intensity or scanner biases; if the latter, the adaptive mechanism reduces to a domain-level average prompt.

Authors: We appreciate the referee highlighting the importance of validating what the low-frequency Fourier features capture. The LFC loss is specifically designed to produce domain-discriminative representations by clustering same-domain samples and separating different-domain ones, and the consistent gains on unseen domains in our experiments indicate that the features support input-specific adaptation beyond global biases. To provide direct evidence, we will revise §3.2 and the experiments section to include t-SNE visualizations of the domain features (colored by domain and acquisition parameters), nearest-neighbor retrieval examples from the prompt memory for intra- and inter-domain inputs, and qualitative failure-case comparisons between adaptive and static prompts. These additions will clarify that the querying mechanism adapts to relevant anatomical and acquisition variations. revision: yes
Referee: [§4] §4 (experimental validation): The reported performance improvements on unseen domains are described at a high level without statistical significance tests (e.g., paired t-tests or Wilcoxon tests across multiple random seeds), full ablation tables isolating the LFC loss and prompt memory size, or exhaustive baseline comparisons that control for backbone capacity and standard prompting variants. This weakens support for the claim that gains are attributable to the adaptive querying rather than the backbone or generic prompting.

Authors: We agree that additional statistical rigor and controlled ablations would better isolate the contribution of the adaptive mechanism. In the revised §4, we will add paired t-tests (and Wilcoxon signed-rank tests where appropriate) computed over multiple random seeds for all reported improvements on unseen domains; complete ablation tables varying the LFC loss weight and prompt memory sizes; and expanded baseline comparisons that include standard visual prompting variants while matching backbone capacities and parameter counts. These revisions will strengthen the attribution of gains to our input-specific querying approach. revision: yes

Circularity Check

0 steps flagged

No significant circularity in APEX derivation chain

full rationale

The paper introduces APEX as a new framework with independent components (learnable prompt memory queried by Fourier-spectrum domain features, plus LFC contrastive loss for domain clustering). These elements are defined and motivated directly in the text without reducing by construction to fitted parameters, self-referential equations, or prior self-citations that bear the central load. Generalization claims rest on empirical results across seen/unseen domains rather than on any mathematical equivalence to inputs. No self-definitional, fitted-input-as-prediction, or ansatz-smuggling patterns appear in the abstract or method description. This is a standard non-circular proposal of a plug-and-play method.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 2 invented entities

The framework depends on the assumption that low-frequency Fourier features discriminate domains and on several hyperparameters for the memory and contrastive loss; new constructs include the prompt memory and LFC learning.

free parameters (2)

Prompt memory size
The capacity of the learnable prompt memory is a tunable hyperparameter required for the retrieval mechanism.
LFC loss weighting coefficients
Weights balancing the contrastive terms for clustering same-domain and separating different-domain features must be chosen or fitted.

axioms (1)

domain assumption Low-frequency components of the Fourier spectrum contain sufficient domain-discriminative information for medical images
Invoked when the method extracts domain features from the Fourier spectrum to query the prompt memory.

invented entities (2)

Learnable prompt memory no independent evidence
purpose: Stores diverse domain-discriminative prompt representations for input-specific retrieval
New data structure introduced to enable adaptive rather than uniform prompting.
Low-Frequency Feature Contrastive (LFC) learning framework no independent evidence
purpose: Trains domain features to cluster same-domain samples and separate different-domain samples
Novel training procedure proposed to learn robust domain features.

pith-pipeline@v0.9.0 · 5506 in / 1401 out tokens · 48976 ms · 2026-05-10T05:24:05.733214+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

50 extracted references

[1]

Medical image segmentation review: The suc- cess of u-net.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024

Reza Azad, Ehsan Khodapanah Aghdam, Amelie Rauland, Yiwei Jia, Atlas Haddadi Avval, Afshin Bozorgpour, Sanaz Karimijafarbigloo, Joseph Paul Cohen, Ehsan Adeli, and Dorit Merhof. Medical image segmentation review: The suc- cess of u-net.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024. 1

2024
[2]

Wm-dova maps for accurate polyp highlighting in colonoscopy: Validation vs

Jorge Bernal, F Javier S ´anchez, Gloria Fern ´andez- Esparrach, Debora Gil, Cristina Rodr ´ıguez, and Fernando Vilari˜no. Wm-dova maps for accurate polyp highlighting in colonoscopy: Validation vs. saliency maps from physi- cians.Computerized medical imaging and graphics, 43:99– 111, 2015. 5

2015
[3]

Style mixup enhanced disentanglement learning for unsupervised domain adaptation in medical im- age segmentation.Medical Image Analysis, 101:103440,

Zhuotong Cai, Jingmin Xin, Chenyu You, Peiwen Shi, Siyuan Dong, Nicha C Dvornek, Nanning Zheng, and James S Duncan. Style mixup enhanced disentanglement learning for unsupervised domain adaptation in medical im- age segmentation.Medical Image Analysis, 101:103440,
[4]

Swin-unet: Unet-like pure transformer for medical image segmentation,

Hu Cao, Yueyue Wang, Joy Chen, Dongsheng Jiang, Xi- aopeng Zhang, Qi Tian, and Manning Wang. Swin-unet: Unet-like pure transformer for medical image segmentation,
[5]

Transunet: Rethinking the u-net architec- ture design for medical image segmentation through the lens of transformers.Medical Image Analysis, 97:103280, 2024

Jieneng Chen, Jieru Mei, Xianhang Li, Yongyi Lu, Qihang Yu, Qingyue Wei, Xiangde Luo, Yutong Xie, Ehsan Adeli, Yan Wang, et al. Transunet: Rethinking the u-net architec- ture design for medical image segmentation through the lens of transformers.Medical Image Analysis, 97:103280, 2024. 5

2024
[6]

Each test image deserves a specific prompt: Con- tinual test-time adaptation for 2d medical image segmenta- tion

Ziyang Chen, Yongsheng Pan, Yiwen Ye, Mengkang Lu, and Yong Xia. Each test image deserves a specific prompt: Con- tinual test-time adaptation for 2d medical image segmenta- tion. InProceedings of the IEEE/CVF conference on com- puter vision and pattern recognition, pages 11184–11193,
[7]

Complemen- tary domain adaptation and generalization for unsupervised continual domain shift learning

Wonguk Cho, Jinha Park, and Taesup Kim. Complemen- tary domain adaptation and generalization for unsupervised continual domain shift learning. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 11442–11452, 2023. 1, 2

2023
[8]

Sailesh Conjeti, Amin Katouzian, Abhijit Guha Roy, Lo ¨ıc Peter, Debdoot Sheet, Stephane Carlier, Andrew Laine, and Nassir Navab. Supervised domain adaptation of decision forests: Transfer of models trained in vitro for in vivo in- travascular ultrasound tissue characterization.Medical im- age analysis, 32:1–17, 2016. 2

2016
[10]

Towards accurate cardiac mri segmenta- tion with variational autoencoder-based unsupervised do- main adaptation.IEEE Transactions on Medical Imaging,

Hengfei Cui, Yan Li, Yifan Wang, Di Xu, Lian-Ming Wu, and Yong Xia. Towards accurate cardiac mri segmenta- tion with variational autoencoder-based unsupervised do- main adaptation.IEEE Transactions on Medical Imaging,
[11]

Pranet: Parallel reverse attention network for polyp segmentation

Deng-Ping Fan, Ge-Peng Ji, Tao Zhou, Geng Chen, Huazhu Fu, Jianbing Shen, and Ling Shao. Pranet: Parallel reverse attention network for polyp segmentation. InInternational conference on medical image computing and computer- assisted intervention, pages 263–273. Springer, 2020. 5

2020
[12]

Exploring the clinical potential of an automatic colonic polyp detection method based on the creation of energy maps.Endoscopy, 48(09):837–842, 2016

Gl `oria Fern ´andez-Esparrach, Jorge Bernal, Maria L ´opez- Cer´on, Henry C ´ordova, Cristina S ´anchez-Montes, Cristina Rodr ´ıguez De Miguel, and Francisco Javier S´anchez. Exploring the clinical potential of an automatic colonic polyp detection method based on the creation of energy maps.Endoscopy, 48(09):837–842, 2016. 5

2016
[13]

Rim-one: An open reti- nal image database for optic nerve evaluation

Francisco Fumero, Silvia Alay ´on, Jos ´e L Sanchez, Jose Sigut, and M Gonzalez-Hernandez. Rim-one: An open reti- nal image database for optic nerve evaluation. In2011 24th international symposium on computer-based medical sys- tems (CBMS), pages 1–6. IEEE, 2011. 5

2011
[14]

Transfer learning for domain adaptation in mri: Application in brain lesion segmentation

Mohsen Ghafoorian, Alireza Mehrtash, Tina Kapur, Nico Karssemeijer, Elena Marchiori, Mehran Pesteie, Charles RG Guttmann, Frank-Erik De Leeuw, Clare M Tempany, Bram Van Ginneken, et al. Transfer learning for domain adaptation in mri: Application in brain lesion segmentation. InMedi- cal Image Computing and Computer Assisted Intervention- MICCAI 2017: 20th...

2017
[16]

Supervised domain adaptation by transferring both the pa- rameter set and its gradient.Neurocomputing, 560:126828,

Shaya Goodman, Hayit Greenspan, and Jacob Goldberger. Supervised domain adaptation by transferring both the pa- rameter set and its gradient.Neurocomputing, 560:126828,
[17]

Frogdognet: Fourier frequency retained visual prompt output guidance for domain generalization of clip in remote sensing

Hariseetharam Gunduboina, Muhammad Haris Khan, and Biplab Banerjee. Frogdognet: Fourier frequency retained visual prompt output guidance for domain generalization of clip in remote sensing. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, pages 2384–2397, 2025. 3

2025
[18]

M Hamed Mozaffari and Won-Sook Lee. Domain adaptation for ultrasound tongue contour extraction using transfer learn- ing: A deep learning approach.The Journal of the Acoustical Society of America, 146(5):EL431–EL437, 2019. 2

2019
[19]

Prompting to adapt 9 foundational segmentation models

Jie Hu, Jie Li, Yue Ma, Liujuan Cao, Songan Zhang, Wei Zhang, Guannan Jiang, and Rongrong Ji. Prompting to adapt 9 foundational segmentation models. InProceedings of the 32nd ACM International Conference on Multimedia, pages 7104–7112, 2024. 3

2024
[20]

Context-aware pseudo-label refinement for source-free do- main adaptive fundus image segmentation

Zheang Huai, Xinpeng Ding, Yi Li, and Xiaomeng Li. Context-aware pseudo-label refinement for source-free do- main adaptive fundus image segmentation. InInternational Conference on Medical Image Computing and Computer- Assisted Intervention, pages 618–628. Springer, 2023. 2

2023
[21]

Kvasir-seg: A segmented polyp dataset

Debesh Jha, Pia H Smedsrud, Michael A Riegler, P ˚al Halvorsen, Thomas De Lange, Dag Johansen, and H˚avard D Johansen. Kvasir-seg: A segmented polyp dataset. In MultiMedia modeling: 26th international conference, MMM 2020, Daejeon, South Korea, January 5–8, 2020, proceed- ings, part II 26, pages 451–462. Springer, 2020. 5

2020
[22]

Vi- sual prompt tuning

Menglin Jia, Luming Tang, Bor-Chun Chen, Claire Cardie, Serge Belongie, Bharath Hariharan, and Ser-Nam Lim. Vi- sual prompt tuning. InEuropean conference on computer vision, pages 709–727. Springer, 2022. 1, 3, 5

2022
[23]

Prompt tuning of deep neural networks for speaker-adaptive visual speech recognition.IEEE Transactions on Pattern Analysis & Machine Intelligence, 47(02):1042–1055, 2025

Minsu Kim, Hyung-Il Kim, and Yong Man Ro. Prompt tuning of deep neural networks for speaker-adaptive visual speech recognition.IEEE Transactions on Pattern Analysis & Machine Intelligence, 47(02):1042–1055, 2025. 1, 3, 5

2025
[24]

Segment any- thing

Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer White- head, Alexander C Berg, Wan-Yen Lo, et al. Segment any- thing. InProceedings of the IEEE/CVF international confer- ence on computer vision, pages 4015–4026, 2023. 3

2023
[25]

Structure boundary preserving segmen- tation for medical image with ambiguous boundary

Hong Joo Lee, Jung Uk Kim, Sangmin Lee, Hak Gu Kim, and Yong Man Ro. Structure boundary preserving segmen- tation for medical image with ambiguous boundary. InPro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4817–4826, 2020. 1

2020
[26]

Unsupervised domain adaptation for medical image segmentation using adaptogen-perturbation.Medical Image Analysis, page 104002, 2026

Hong Joo Lee, Yuan Bi, Sangmin Lee, Gyeong-Moon Park, Jung Uk Kim, Seong Tae Kim, Zhongliang Jiang, and Nassir Navab. Unsupervised domain adaptation for medical image segmentation using adaptogen-perturbation.Medical Image Analysis, page 104002, 2026. 1

2026
[27]

Video prediction recalling long-term mo- tion context via memory alignment learning

Sangmin Lee, Hak Gu Kim, Dae Hwi Choi, Hyung-Il Kim, and Yong Man Ro. Video prediction recalling long-term mo- tion context via memory alignment learning. InProceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, pages 3054–3063, 2021. 4

2021
[28]

Weakly paired associative learning for sound and image represen- tations via bimodal associative memory

Sangmin Lee, Hyung-Il Kim, and Yong Man Ro. Weakly paired associative learning for sound and image represen- tations via bimodal associative memory. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10534–10543, 2022. 4

2022
[29]

Pre-train, prompt, and predict: A systematic survey of prompting methods in nat- ural language processing.ACM computing surveys, 55(9): 1–35, 2023

Pengfei Liu, Weizhe Yuan, Jinlan Fu, Zhengbao Jiang, Hi- roaki Hayashi, and Graham Neubig. Pre-train, prompt, and predict: A systematic survey of prompting methods in nat- ural language processing.ACM computing surveys, 55(9): 1–35, 2023. 3

2023
[30]

Modality-agnostic domain generalizable medical image segmentation by multi-frequency in multi- scale attention

Ju-Hyeon Nam, Nur Suriza Syazwany, Su Jung Kim, and Sang-Chul Lee. Modality-agnostic domain generalizable medical image segmentation by multi-frequency in multi- scale attention. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11480– 11491, 2024. 2, 4

2024
[31]

Neounet: Towards accurate colon polyp segmentation and neoplasm detection

Phan Ngoc Lan, Nguyen Sy An, Dao Viet Hang, Dao Van Long, Tran Quang Trung, Nguyen Thi Thuy, and Dinh Viet Sang. Neounet: Towards accurate colon polyp segmentation and neoplasm detection. InAdvances in visual computing: 16th international symposium, ISVC 2021, virtual event, Oc- tober 4-6, 2021, proceedings, part II, pages 15–28. Springer,

2021
[32]

Refuge challenge: A unified framework for evaluat- ing automated methods for glaucoma assessment from fun- dus photographs.Medical image analysis, 59:101570, 2020

Jos ´e Ignacio Orlando, Huazhu Fu, Jo ˜ao Barbosa Breda, Karel Van Keer, Deepti R Bathula, Andr ´es Diaz-Pinto, Ruogu Fang, Pheng-Ann Heng, Jeyoung Kim, JoonHo Lee, et al. Refuge challenge: A unified framework for evaluat- ing automated methods for glaucoma assessment from fun- dus photographs.Medical image analysis, 59:101570, 2020. 5

2020
[33]

Two at once: Enhancing learning and generalization capacities via ibn-net

Xingang Pan, Ping Luo, Jianping Shi, and Xiaoou Tang. Two at once: Enhancing learning and generalization capacities via ibn-net. InProceedings of the european conference on computer vision (ECCV), pages 464–479, 2018. 1

2018
[34]

Endometrium segmentation on transvaginal ultra- sound image using key-point discriminator.Medical physics, 46(9):3974–3984, 2019

Hyenok Park, Hong Joo Lee, Hak Gu Kim, Yong Man Ro, Dongkuk Shin, Sa Ra Lee, Sung Hoon Kim, and Mikyung Kong. Endometrium segmentation on transvaginal ultra- sound image using key-point discriminator.Medical physics, 46(9):3974–3984, 2019. 1

2019
[35]

Rdumb: A simple approach that questions our progress in continual test-time adaptation.Advances in Neural Information Processing Systems, 36:39915–39935,

Ori Press, Steffen Schneider, Matthias K ¨ummerer, and Matthias Bethge. Rdumb: A simple approach that questions our progress in continual test-time adaptation.Advances in Neural Information Processing Systems, 36:39915–39935,
[36]

Medical image segmentation via single- source domain generalization with random amplitude spec- trum synthesis

Qiang Qiao, Wenyu Wang, Meixia Qu, Kun Su, Bin Jiang, and Qiang Guo. Medical image segmentation via single- source domain generalization with random amplitude spec- trum synthesis. InInternational Conference on Medical Im- age Computing and Computer-Assisted Intervention, pages 435–445. Springer, 2024. 2, 4, 5

2024
[37]

Deep learning for medical image segmentation: State-of-the- art advancements and challenges.Informatics in Medicine Unlocked, page 101504, 2024

Md Eshmam Rayed, SM Sajibul Islam, Sadia Islam Niha, Jamin Rahman Jim, Md Mohsin Kabir, and MF Mridha. Deep learning for medical image segmentation: State-of-the- art advancements and challenges.Informatics in Medicine Unlocked, page 101504, 2024. 1

2024
[38]

U- net: Convolutional networks for biomedical image segmen- tation

Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U- net: Convolutional networks for biomedical image segmen- tation. InInternational Conference on Medical image com- puting and computer-assisted intervention, pages 234–241. Springer, 2015. 5

2015
[39]

Toward embedded detection of polyps in wce images for early diagnosis of colorectal can- cer.International journal of computer assisted radiology and surgery, 9:283–293, 2014

Juan Silva, Aymeric Histace, Olivier Romain, Xavier Dray, and Bertrand Granado. Toward embedded detection of polyps in wce images for early diagnosis of colorectal can- cer.International journal of computer assisted radiology and surgery, 9:283–293, 2014. 5

2014
[40]

Drishti-gs: Reti- nal image dataset for optic nerve head (onh) segmentation

Jayanthi Sivaswamy, SR Krishnadas, Gopal Datt Joshi, Mad- hulika Jain, and A Ujjwaft Syed Tabish. Drishti-gs: Reti- nal image dataset for optic nerve head (onh) segmentation. In2014 IEEE 11th international symposium on biomedical imaging (ISBI), pages 53–56. IEEE, 2014. 5

2014
[41]

Source-free domain adaptive fundus image segmen- tation with class-balanced mean teacher

Longxiang Tang, Kai Li, Chunming He, Yulun Zhang, and Xiu Li. Source-free domain adaptive fundus image segmen- tation with class-balanced mean teacher. InInternational 10 Conference on Medical Image Computing and Computer- Assisted Intervention, pages 684–694. Springer, 2023. 2

2023
[42]

Source-free domain adaptive fundus image segmen- tation with class-balanced mean teacher

Longxiang Tang, Kai Li, Chunming He, Yulun Zhang, and Xiu Li. Source-free domain adaptive fundus image segmen- tation with class-balanced mean teacher. InInternational Conference on Medical Image Computing and Computer- Assisted Intervention, pages 684–694. Springer, 2023. 1

2023
[43]

Fvp: Fourier visual prompting for source-free unsupervised do- main adaptation of medical image segmentation.IEEE Transactions on Medical Imaging, 42(12):3738–3751, 2023

Yan Wang, Jian Cheng, Yixin Chen, Shuai Shao, Lanyun Zhu, Zhenzhou Wu, Tao Liu, and Haogang Zhu. Fvp: Fourier visual prompting for source-free unsupervised do- main adaptation of medical image segmentation.IEEE Transactions on Medical Imaging, 42(12):3738–3751, 2023. 1, 3, 5

2023
[44]

Source-free do- main adaptation for medical image segmentation via selec- tively updated mean teacher

Ziqi Wen, Xinru Zhang, and Chuyang Ye. Source-free do- main adaptation for medical image segmentation via selec- tively updated mean teacher. InInternational Conference on Information Processing in Medical Imaging, pages 225–236. Springer, 2023. 1

2023
[45]

One-prompt to segment all med- ical images

Junde Wu and Min Xu. One-prompt to segment all med- ical images. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11302– 11312, 2024. 3

2024
[46]

Source free domain adaptation for medical image segmenta- tion with fourier style mining.Medical Image Analysis, 79: 102457, 2022

Chen Yang, Xiaoqing Guo, Zhen Chen, and Yixuan Yuan. Source free domain adaptation for medical image segmenta- tion with fourier style mining.Medical Image Analysis, 79: 102457, 2022. 1, 3

2022
[47]

Class-aware feature alignment for domain adaptative mito- chondria segmentation

Dan Yin, Wei Huang, Zhiwei Xiong, and Xuejin Chen. Class-aware feature alignment for domain adaptative mito- chondria segmentation. InInternational Conference on Med- ical Image Computing and Computer-Assisted Intervention, pages 238–248. Springer, 2023. 1

2023
[48]

Class-aware feature alignment for domain adaptative mito- chondria segmentation

Dan Yin, Wei Huang, Zhiwei Xiong, and Xuejin Chen. Class-aware feature alignment for domain adaptative mito- chondria segmentation. InInternational Conference on Med- ical Image Computing and Computer-Assisted Intervention, pages 238–248. Springer, 2023. 2

2023
[49]

A2xp: Towards pri- vate domain generalization

Geunhyeok Yu and Hyoseok Hwang. A2xp: Towards pri- vate domain generalization. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 23544–23553, 2024. 1, 3, 5

2024
[50]

Source-free domain adaptation for medical image segmentation via prototype-anchored feature alignment and contrastive learning

Qinji Yu, Nan Xi, Junsong Yuan, Ziyu Zhou, Kang Dang, and Xiaowei Ding. Source-free domain adaptation for medical image segmentation via prototype-anchored feature alignment and contrastive learning. InInternational Confer- ence on Medical Image Computing and Computer-Assisted Intervention, pages 3–12. Springer, 2023. 2

2023
[51]

Mapseg: Uni- fied unsupervised domain adaptation for heterogeneous med- ical image segmentation based on 3d masked autoencoding and pseudo-labeling

Xuzhe Zhang, Yuhao Wu, Elsa Angelini, Ang Li, Jia Guo, Jerod M Rasmussen, Thomas G O’Connor, Pathik D Wad- hwa, Andrea Parolin Jackowski, Hai Li, et al. Mapseg: Uni- fied unsupervised domain adaptation for heterogeneous med- ical image segmentation based on 3d masked autoencoding and pseudo-labeling. InProceedings of the IEEE/CVF Con- ference on Computer ...

2024
[52]

Road extraction by deep residual u-net.IEEE Geoscience and Re- mote Sensing Letters, 15(5):749–753, 2018

Zhengxin Zhang, Qingjie Liu, and Yunhong Wang. Road extraction by deep residual u-net.IEEE Geoscience and Re- mote Sensing Letters, 15(5):749–753, 2018. 5 11

2018