When Can Test-Time Adaptation Help Zero-Shot CT Vision-Language Models?

Ailar Mahdizadeh; Leonid Sigal; Puria Azadi Moghadam; Xiangteng He

REVIEW 3 major objections 6 minor 24 references

Test-time adaptation helps zero-shot 3D CT vision-language models only when the scan keeps its native depth and the base model already distinguishes present from absent abnormalities; in that regime, the new CARVE method gives consistent mu

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-01 22:57 UTC pith:3RIIM6HN

load-bearing objection A careful, honest empirical study of when TTA helps zero-shot CT VLMs; the main weakness is that CARVE's cardinality estimator (Eq. 8) is never directly validated. the 3 major comments →

arxiv 2607.15556 v1 pith:3RIIM6HN submitted 2026-07-17 cs.CV

When Can Test-Time Adaptation Help Zero-Shot CT Vision-Language Models?

Ailar Mahdizadeh , Puria Azadi Moghadam , Xiangteng He , Leonid Sigal This is my paper

classification cs.CV

keywords test-time adaptationzero-shot3D CTvision-language modelmulti-label classificationentropy minimizationcardinality estimationdistribution shift

verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Test-time adaptation is not automatically useful for zero-shot 3D CT vision-language models. The paper shows two conditions must hold: the input volume must preserve the depth structure the encoder was pretrained on (cutting depth alone dropped internal AUROC by more than 0.12), and the base model's prompt-pair probabilities must already separate present from absent abnormalities. When those conditions hold, standard entropy objectives are structurally mismatched to the multi-label prompt-pair setup, so the paper introduces CARVE, which estimates how many abnormalities a scan actually has and sharpens exactly that many labels. In experiments, CARVE gives the most consistent AUROC improvements across backbones and label settings when the base is discriminative, while no objective can create discrimination when the base representation is near chance.

Core claim

The paper's central claim is that TTA helps zero-shot 3D CT VLMs only in a specific regime: when the volumetric input matches the encoder's pretraining depth and the base prompt-pair probabilities already separate present from absent abnormalities. Within that regime, the correct adaptation objective is cardinality-aware. CARVE estimates the scan-specific number of positive abnormalities k̂ from the sum of per-label Bernoulli probabilities, then optimizes a top-k̂ objective that sharpens the estimated positive set while suppressing the rest, using only the most confident low-entropy weak views for memory-efficient gradient updates. The paper demonstrates that this beats entropy-based baselin

What carries the argument

The central object is the prompt-pair Bernoulli probability p_j = σ(s_j^+ − s_j^-), obtained by a two-class softmax over the positive and negative text-prompt scores for each abnormality. CARVE's mechanism: estimate the number of present abnormalities as k̂ = clip(round(Σ_j p̄_j), k_min, k_max) from the retained-view average; then optimize L_CARVE = −(1/|P|)Σ_{j∈P} log p_j − λ_neg (1/|N|)Σ_{j∈N} log(1−p_j), where P is the top-k̂ label set and N its complement. This replaces standard entropy minimization, which collapses co-occurring positives, with an objective that preserves exactly k̂ abnormalities. View selection scores stochastic weak 3D views without gradients and retains the K lowest-e

Load-bearing premise

The method assumes the sum of the zero-shot per-label probabilities is a calibrated estimate of the true number of abnormalities in the scan, so rounding that sum gives the right positive-label count to sharpen.

What would settle it

Compute the per-scan cardinality estimate k̂ against the true number of abnormalities on an external cohort; if the mean absolute error is large or if CARVE's AUROC gain disappears when k̂ is replaced by the true count, the central claim fails.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

If CARVE's conditions hold, zero-shot CT VLMs can be adapted per-scan without labels, making cross-institution deployment more reliable when the base is already discriminative.
Depth reduction (e.g., z=40 from z=240) is not benign: it can hurt AUROC by >0.12 before adaptation, so deployment pipelines must preserve native depth.
CARVE transfers to smaller label spaces (three-class and binary) without modification because the cardinality estimate reduces naturally.
In the paper's experiments, when the base is near chance, adaptation only shifts the operating point, so effort should go to improving base transfer rather than TTA.
The dispersion gate is not a reliable per-scan viability detector; it only abstains when predictions collapse, so it does not rescue adaptation in the non-discriminative regime.

Where Pith is reading between the lines

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

The cardinality estimate in Eq. 8 assumes the zero-shot probabilities are calibrated enough that their sum approximates the true label count; if miscalibration is large under distribution shift, the top-k̂ objective will sharpen an incorrect positive set. A direct calibration check on external cohorts would test this.
The paper's framing implies a deployment rule: first measure base discriminability on a small labeled sample (or via dispersion) and only then decide whether to run TTA. This could be formalized into a decision criterion with a threshold on base separation.
Since the base model gap dominates the adapter gap under external shift, annotations spent on adapting the base representation (e.g., few-shot fine-tuning) may pay off more than any TTA objective; this is an implicit priority for practitioners.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Desk Editor's Note

A careful, honest empirical study of when TTA helps zero-shot CT VLMs; the main weakness is that CARVE's cardinality estimator (Eq. 8) is never directly validated.

read the letter

Two things you should know. First, this is one of the few TTA papers that seriously asks when adaptation helps rather than assuming it always does, and the diagnostic part is genuinely useful. Second, the method's core mechanism—estimating the positive-label count from the model's own probabilities—is plausible but never directly validated, and the central claim leans on it more than the experiments justify.

What's actually new: CARVE is the first TTA method designed for zero-shot multi-label 3D CT VLMs with prompt-pair Bernoulli probabilities. The memory-efficient retained-view adaptation is a practical contribution for volumetric inference. The empirical trends are consistent: CARVE improves AUROC by about 0.03–0.05 across CT-RATE, RAD-ChestCT, CC-CCII, and LUNA16 on both CT-CLIP and fVLM, while TENT, RLCF, and ML-TTA mostly produce near-zero or inconsistent gains. The paper is also honest about its limits: it explicitly says the dispersion gate is not a per-scan viability detector, and that no objective fixes a near-chance base model. That honesty is real and welcome.

The soft spots. Eq. 8 estimates k-hat as the rounded sum of per-label Bernoulli probabilities. That assumes those probabilities are additively calibrated—their sum approximates the true label count. AUROC, which the paper uses to define "the base model is already discriminative," is order-based, so a model can rank present vs. absent perfectly while all probabilities sit near 0.5, making k-hat roughly L/2. The paper never validates k-hat against true counts and never ablates it. Figure 6 shows the same probabilities carry little per-scan viability signal. So the "cardinality-aware" part of CARVE is unsecured; the reported gains could in principle come from the retained-view selection or the norm-only update. That's addressable—an ablation with true k or a fixed k would settle it—but as written it's a real gap. Also, all reported variability is seed-level sigma, not patient-level confidence intervals, so the effect sizes look more certain than they are. The code link has no commit hash, a minor reproducibility annoyance.

This paper is for anyone working on TTA or medical VLMs. It deserves a serious referee: the question is important, the diagnostic is well-designed, and the weaknesses are fixable rather than structural. Send it to review; ask the authors to either validate the cardinality estimate or show that CARVE works without it.

Referee Report

3 major / 6 minor

Summary. The paper studies when test-time adaptation (TTA) helps zero-shot 3D CT vision-language models (VLMs). Through a controlled diagnostic protocol with two backbones (CT-CLIP, fVLM) and four datasets (CT-RATE, RAD-ChestCT, CC-CCII, LUNA16), it identifies preconditions for TTA: the input pipeline must preserve the volumetric depth structure expected by the encoder, and the base representation must transfer to the target cohort. For the regime where the base model is already discriminative, the paper proposes CARVE, which estimates a sample-specific positive-label cardinality, optimizes a top-k objective on a set of retained weak 3D views, and updates only visual normalization parameters in an episodic manner. The main empirical claim is that CARVE provides consistent AUROC improvements of roughly 0.03–0.05 over no adaptation and over existing TTA baselines (TENT, RLCF, ML-TTA) when the base model already separates present from absent abnormalities, while the baselines show near-zero or inconsistent gains.

Significance. If the findings hold, the paper makes a useful contribution: it is, to my knowledge, the first TTA method designed for multi-label zero-shot 3D CT diagnosis, and it clearly delineates regimes where TTA helps or does not. The diagnostic study is careful in isolating depth structure and base representational transfer, and the authors are transparent about limitations (e.g., the dispersion gate in Fig. 6 is explicitly shown to be a weak per-scan viability signal). The code is released, and the protocol controls for label shift by fixing hyperparameters on the internal validation set. However, the central mechanism—the sample-specific cardinality estimate of Eq. (8)—is not directly validated, and the reported statistical evidence is based on seed-level variability rather than patient-level uncertainty. These issues are load-bearing for the central claim, so the paper cannot be accepted in its current form.

major comments (3)

[Sec. 4.4, Eq. (8)] The cardinality estimate k_hat = clip(round(sum_j pbar_j), k_min, k_max) assumes that the sum of per-label Bernoulli probabilities approximates the true positive-label count (additive calibration). AUROC discriminability is order-based and does not imply such calibration; a model can have high AUROC while all probabilities sit near 0.5, yielding k_hat near L/2. Under external distribution shift, miscalibration is likely, and Fig. 6 shows that the same probabilities carry little per-scan viability signal. If k_hat is wrong, the top-k index set in Eq. (9) can exclude true positives, and the negative loss term actively suppresses them. The paper never validates k_hat against true counts nor ablates Eq. (8) (e.g., comparing fixed k, oracle k, or a recalibrated estimator). Since the central claim attributes CARVE's gains to cardinality-aware adaptation, this mechanism is load-bearing and curr
[Sec. 5.1 / Appendix B] The reported ±σ is over adaptation seeds, not patients. The authors explicitly state in Appendix B that the standard deviation reflects run-to-run stability under stochastic inference-time distortions and is "not statistical uncertainty over patients." AUROC improvements are about 0.03–0.05, while seed-level σ values are often 0.0001–0.003. Without patient-level bootstrap confidence intervals or an appropriate significance test across patients, the claim that CARVE "provides the most consistent improvements" is not statistically grounded. This is a load-bearing issue because the central conclusion rests on the consistency of small-magnitude gains across cohorts.
[Sec. 5.1, Methods and protocol] The implementation of the baselines RLCF and ML-TTA is not specified. ML-TTA [19] in its original form relies on caption retrieval for pseudo-label filtering; zero-shot CT scans have no paired captions. If ML-TTA is run without its caption component, it is not the published method, and the comparison may be unfair; if captions are simulated or omitted, that should be stated. Similarly, RLCF [23] uses a CLIP reward whose exact computation for CT-CLIP/fVLM is not described. Without these details, the experimental comparison is not reproducible, undermining the conclusion that CARVE outperforms strong baselines.

minor comments (6)

[Sec. 5.3, Fig. 2] The hyperparameter sensitivity sweep is performed post-hoc on the external RAD-ChestCT test set. The authors state it is not used for selection, but this is still an exploration of a test cohort and should be acknowledged more strongly, e.g., as a limitation on the strength of the robustness claim.
[Appendix B, fVLM evaluation] The fVLM experiments omit the official organ-ROI segmentation crop. The authors disclose this omission, which is good, but the resulting numbers should be presented as an approximation rather than a direct comparison to published fVLM results.
[Sec. 4.4, Eq. (8)] The choice of k_min=1 and k_max=8 is reported in Appendix A but not justified in the main text. It would be helpful to explain whether the upper bound is derived from the label-space size, a prevalence prior, or an empirical heuristic, since this directly affects the behavior of Eq. (8).
[Abstract / Introduction] The phrase "the first TTA method for this setting" is strong; suggest softening to "to our knowledge, the first" to hedge against possible concurrent work.
[Throughout] Figure captions are dense, especially Fig. 4, which mixes two failure axes, gradient magnitude and direction. Expanding the caption or adding a short text explanation in Sec. 5.3 would improve readability.
[Sec. 4.5, Eq. (10)] Minor typo: "We minimize it over the adapted normalization parameters φ and then predict on the original unperturbed volume." The sentence is missing a period after "volume." Also, the equation label "L_TTA" is referred to inconsistently in the text.

Circularity Check

0 steps flagged

No significant circularity: CARVE's self-referential pseudo-labeling is a method design, not a derivation, and all headline claims are benchmark-tested.

full rationale

CARVE does contain a self-referential loop: Eq. 8 estimates the cardinality k̂ from the model's own averaged probabilities (k̂ = clip(round(Σ_j p̄_j), k_min, k_max)) and Eq. 9 defines the positive set as TopK_{k̂}(p^(v)), so the pseudo-labels being sharpened are the model's current most-confident labels. This is a standard self-training-style objective and is not presented as a derivation of ground truth; the paper explicitly calls k̂ an 'estimate' and P^(v) the 'likely positive labels.' The paper's headline claims are not proven by this loop but by external benchmark comparisons (Tables 1–5) against No-TTA, TENT, RLCF, and ML-TTA on ground-truth labels, and the diagnostic 'when' conditions (depth integrity, base transfer) are established by controlled AUROC comparisons that do not depend on the CARVE objective. There are no load-bearing self-citations: the cited backbones (CT-CLIP, fVLM) and TTA baselines are external prior work, and no uniqueness theorem from the authors' own papers is invoked. The paper also states its own limitations (Fig. 6: σ_pred is 'near random' and not a per-scan viability detector; Appendix B: no parity claimed with fVLM's reported numbers), which further weighs against a hidden circular derivation. The unvalidated cardinality estimate in Eq. 8 is a correctness/calibration risk, not a circularity: failing to validate an auxiliary estimate is not the same as assuming the conclusion.

Axiom & Free-Parameter Ledger

6 free parameters · 4 axioms · 0 invented entities

The central method rests on a handful of chosen hyperparameters and on domain assumptions about Bernoulli prompt-pair probabilities, entropy-based view reliability, cardinality estimation, and norm-only episodic updates. No new physical entities are introduced, and no derivation-level circularity is present.

free parameters (6)

V (number of weak views) = 8
Fixed across all experiments and selected on CT-RATE validation; sensitivity shown in Fig. 2.
rho (view retention ratio) = 0.25
Retains 25% of lowest-entropy views; fixed across all cohorts and selected on internal validation.
lambda_neg (negative loss weight) = 0.8
Weight for suppression of the complement set in Eq. 9; fixed and reused unchanged.
k_min, k_max (cardinality bounds) = 1, 8
Clip bounds for Eq. 8; chosen to avoid empty positive sets and implausibly large positive sets; fixed across all cohorts.
tau (dispersion gate threshold) = 0.091
Selected on internal validation split; used only as a conservative abstention signal.
learning rate and steps = 1e-5, 2 steps
Adaptation schedule fixed for all experiments; selected on internal validation.

axioms (4)

domain assumption Prompt-pair softmax defines independent Bernoulli probabilities for each abnormality (Eq. 3).
The whole CARVE objective treats labels as independent given the prompt pair; label correlations or miscalibration would break the top-k logic.
domain assumption Per-label entropy is a reliable signal for selecting augmented views (Eq. 6-7).
CARVE retains low-entropy views; the paper tests perturbation strength but does not directly verify that entropy selects views whose gradient direction is correct.
domain assumption The sum of Bernoulli probabilities estimates the scan-specific cardinality (Eq. 8).
This is the load-bearing assumption: if probabilities are miscalibrated under shift, k-hat is wrong and the top-k objective sharpens an incorrect positive set.
domain assumption Norm-only episodic adaptation can improve a zero-shot VLM without drift (following Tent [17]).
Adopted from prior work; the paper provides no drift or stability analysis for longer streams.

pith-pipeline@v1.3.0-alltime-deepseek · 18541 in / 13871 out tokens · 150191 ms · 2026-08-01T22:57:18.940056+00:00 · methodology

0 comments

read the original abstract

3D CT vision-language models (VLMs) classify abnormalities from text prompts in a zero-shot manner, enabling cross-institution deployment where labels are scarce and clinical tasks shift faster than supervised models can be retrained. A real CT scan, however, typically contains several co-occurring abnormalities, and the reliability of zero-shot multi-label prediction under distribution shift remains poorly understood. Test-time adaptation (TTA) updates a model on unlabeled target scans without source data or target annotations, yet existing TTA methods target multi-class softmax prediction on natural images or 2D medical segmentation, and none addresses unsupervised multi-label adaptation for zero-shot 3D CT VLMs. We study when TTA helps zero-shot 3D CT VLMs. A controlled diagnostic analysis shows that TTA is conditional: the volumetric input must preserve the encoder's depth structure, and the base representation must transfer to the target cohort, with depth reduction alone lowering internal AUROC by more than 0.12. We then focus on the regime where the base model already separates present from absent abnormalities. We introduce CARVE (Cardinality-Aware Retained-View Entropy), the first TTA method for this setting. CARVE estimates a sample-specific positive-label cardinality $\hat{k}$, optimizes a top-$\hat{k}$ objective to preserve co-occurring abnormalities, and performs memory-efficient multi-view adaptation by scoring weak 3D views without gradients before updating on a retained subset. Across contrastive CT-CLIP and anatomy-aware fVLM, CARVE provides the most consistent improvements across multi-label, three-class, and binary CT tasks when the base model is already discriminative. These results establish multi-label TTA for zero-shot 3D CT VLMs as a distinct problem and CARVE as a cardinality-aware solution.

Figures

Figures reproduced from arXiv: 2607.15556 by Ailar Mahdizadeh, Leonid Sigal, Puria Azadi Moghadam, Xiangteng He.

**Figure 1.** Figure 1: CARVE overview. Given a base 3D CT VLM (a Medical Text Encoder paired with a 3D vision encoder), each abnormality is scored with positive/negative prompts, yielding independent per-label probabilities. CARVE generates weak volumetric views, retains a reliable low-entropy subset (V ∗ of V views), estimates a sample-specific positive-label cardinality kˆ, and updates only the visual normalization parameters… view at source ↗

**Figure 2.** Figure 2: Hyperparameter sensitivity. CT-CLIP on RADChestCT. CARVE is stable around the default setting V =8, ρ=0.25, and λneg=0.8 [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗

**Figure 3.** Figure 3: Weak view strength. Stronger perturbations raise disagreement but do not improve adaptation, motivating milder distortions. objective can recover discrimination when the base lacks it. 5.3. Ablations and Analysis Component sensitivity [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

**Figure 6.** Figure 6: Risk coverage of the dispersion gate. σpred falls far short of the oracle; the gate only abstains under collapsed predictions to avoid wasteful updates, not detecting per-scan viability. mentation strength increases cross view disagreement, but does not improve agreement with the useful update direction. CARVE therefore uses mild HU jitter and small spatial shifts: the multi view step is for reliability … view at source ↗

discussion (0)

Reference graph

Works this paper leans on

24 extracted references · 2 linked inside Pith

[1]

Shah, Greg Zaharchuk, Marc Willis, Adam Yala, Andrew Johnston, Robert D

Louis Blankemeier, Ashwin Kumar, Joseph Paul Cohen, Jiaming Liu, Longchao Liu, Dave Van Veen, Syed Ja- mal Safdar Gardezi, Hongkun Yu, Magdalini Paschali, Zhi- hong Chen, Jean-Benoit Delbrouck, Eduardo Reis, Rob- bie Holland, Cesar Truyts, Christian Bluethgen, Yufu Wu, Long Lian, Malte Engmann Kjeldskov Jensen, Sophie Ost- meier, Maya Varma, Jeya Maria Jo...

2026
[2]

Gradient align- ment improves test-time adaptation for medical image seg- mentation

Zhe Chen, Yu Ye, Yulin Pan, and Yong Xia. Gradient align- ment improves test-time adaptation for medical image seg- mentation. InProceedings of the AAAI Conference on Arti- ficial Intelligence, pages 2429–2437, 2025. 1, 2

2025
[3]

Diverse data augmentation with diffusions for effective test-time prompt tuning

Chun-Mei Feng, Kai Yu, Yong Liu, Salman Khan, and Wangmeng Zuo. Diverse data augmentation with diffusions for effective test-time prompt tuning. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2023. 2

2023
[4]

Domain-adversarial train- ing of neural networks.Journal of Machine Learning Re- search, 17(59):1–35, 2016

Yaroslav Ganin, Evgeniya Ustinova, Hana Ajakan, Pas- cal Germain, Hugo Larochelle, Franc ¸ois Laviolette, Mario Marchand, and Victor Lempitsky. Domain-adversarial train- ing of neural networks.Journal of Machine Learning Re- search, 17(59):1–35, 2016. 1

2016
[5]

Generalist foundation models from a multimodal dataset for 3d computed tomography.arXiv preprint arXiv:2403.17834,

Ibrahim Ethem Hamamci, Sezgin Er, Chenyu Wang, Furkan Almas, Ayse Gulnihan Simsek, Sevval Nil Esir- gun, Irem Dogan, Omer Faruk Durugol, Benjamin Hou, Suprosanna Shit, Weicheng Dai, Murong Xu, Hadrien Rey- naud, Muhammed Furkan Dasdelen, Bastian Wittmann, Tamaz Amiranashvili, Enis Simsar, Mehmet Simsar, Em- ine Bensu Erdemir, Abdullah Alanbay, Anjany Seku...

arXiv
[6]

Efficient test-time adaptation of vision-language models

Adilbek Karmanov, Dayan Guan, Shijian Lu, Abdulmotaleb El Saddik, and Eric Xing. Efficient test-time adaptation of vision-language models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024. 2

2024
[7]

Do we really need to access the source data? source hypothesis transfer for un- supervised domain adaptation

Jian Liang, Dapeng Hu, and Jiashi Feng. Do we really need to access the source data? source hypothesis transfer for un- supervised domain adaptation. InProceedings of the 37th In- ternational Conference on Machine Learning, pages 6028– 6039, 2020. 1

2020
[8]

Ct-glip: 3d grounded language-image pretraining with ct scans and ra- diology reports for full-body scenarios.arXiv preprint arXiv:2404.15272, 2024

Jingyang Lin, Yingda Xia, Jianpeng Zhang, Ke Yan, Kai Cao, Le Lu, Jiebo Luo, and Ling Zhang. Ct-glip: 3d grounded language-image pretraining with ct scans and ra- diology reports for full-body scenarios.arXiv preprint arXiv:2404.15272, 2024. 2

arXiv 2024
[9]

Test-time adaptation with calibration of medical image classification nets for label distribution shift

Xu Ma et al. Test-time adaptation with calibration of medical image classification nets for label distribution shift. InMed- ical Image Computing and Computer Assisted Intervention (MICCAI), 2022. 2

2022
[10]

Efficient test-time model adaptation without forgetting

Shuaicheng Niu, Jiaxiang Wu, Yifan Zhang, Yaofo Chen, Shijian Zheng, Peilin Zhao, and Mingkui Tan. Efficient test-time model adaptation without forgetting. InInterna- tional conference on machine learning, pages 16888–16905. PMLR, 2022. 2

2022
[11]

Towards stable test-time adaptation in dynamic wild world.arXiv preprint arXiv:2302.12400, 2023

Shuaicheng Niu, Jiaxiang Wu, Yifan Zhang, Zhiquan Wen, Yaofo Chen, Peilin Zhao, and Mingkui Tan. Towards stable test-time adaptation in dynamic wild world.arXiv preprint arXiv:2302.12400, 2023. 2

Pith/arXiv arXiv 2023
[12]

Learning transferable visual models from natural language supervision

Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. Learning transferable visual models from natural language supervision. InProceedings of the 38th International Conference on Machine Learning, pages 8748–8763, 2021. 1

2021
[13]

Auto- matic correction of performance drift under acquisition shift in medical image classification.Nature Communications, 14 (1):6608, 2023

M ´elanie Roschewitz, Galvin Khara, Joe Yearsley, Nisha Sharma, Jonathan J James, ´Eva Ambr ´ozay, Adam Heroux, Peter Kecskemethy, Tobias Rijken, and Ben Glocker. Auto- matic correction of performance drift under acquisition shift in medical image classification.Nature Communications, 14 (1):6608, 2023. 1

2023
[14]

Test- time prompt tuning for zero-shot generalization in vision- language models

Manli Shu, Weili Nie, De-An Huang, Zhiding Yu, Tom Goldstein, Anima Anandkumar, and Chaowei Xiao. Test- time prompt tuning for zero-shot generalization in vision- language models. InAdvances in Neural Information Pro- cessing Systems, 2022. 1, 2

2022
[15]

Large-scale and fine-grained vision-language pre-training for enhanced ct image under- standing.arXiv preprint arXiv:2501.14548, 2025

Zhongyi Shui, Jianpeng Zhang, Weiwei Cao, Sinuo Wang, Ruizhe Guo, Le Lu, Lin Yang, Xianghua Ye, Tingbo Liang, Qi Zhang, and Ling Zhang. Large-scale and fine-grained vision-language pre-training for enhanced ct image under- standing.arXiv preprint arXiv:2501.14548, 2025. 1, 2, 5

Pith/arXiv arXiv 2025
[16]

Efros, and Moritz Hardt

Yu Sun, Xiaolong Wang, Zhuang Liu, John Miller, Alexei A. Efros, and Moritz Hardt. Test-time training with self- supervision for generalization under distribution shifts. In Proceedings of the 37th International Conference on Ma- chine Learning, pages 9229–9248, 2020. 1

2020
[17]

Tent: Fully test-time adaptation by entropy minimization

Dequan Wang, Evan Shelhamer, Shaoteng Liu, Bruno Ol- shausen, and Trevor Darrell. Tent: Fully test-time adaptation by entropy minimization. InInternational Conference on Learning Representations (ICLR), 2021. 1, 2, 3, 5, 11

2021
[18]

Continual test-time domain adaptation

Qin Wang, Olga Fink, Luc Van Gool, and Dengxin Dai. Continual test-time domain adaptation. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 7201–7211, 2022. 2

2022
[19]

Multi-label test-time adaptation with bound en- tropy minimization

Xiangyu Wu, Feng Yu, Yang Yang, Qing-Guo Chen, and Jianfeng Lu. Multi-label test-time adaptation with bound en- tropy minimization. InInternational Conference on Learn- ing Representations (ICLR), 2025. 1, 2, 3, 4, 5

2025
[20]

External vali- dation of deep learning algorithms for radiologic diagnosis: a systematic review.Radiology: Artificial Intelligence, 4(3): e210064, 2022

Alice C Yu, Bahram Mohajer, and John Eng. External vali- dation of deep learning algorithms for radiologic diagnosis: a systematic review.Radiology: Artificial Intelligence, 4(3): e210064, 2022. 1

2022
[21]

A large- scale benchmark for test-time adaptation methods in medi- cal image segmentation.arXiv preprint arXiv:2512.02497,

Wenjing Yu, Shuo Jiang, Yifei Chen, Shuo Chang, Yuanhan Wang, Beining Wu, Jie Dong, Mingxuan Liu, Shenghao Zhu, Feiwei Qin, Changmiao Wang, and Qiyuan Tian. A large- scale benchmark for test-time adaptation methods in medi- cal image segmentation.arXiv preprint arXiv:2512.02497,

arXiv
[22]

Realistic test-time adaptation of vision-language models

Maxime Zanella, Cl ´ement Fuchs, Christophe De Vleeschouwer, and Ismail Ben Ayed. Realistic test-time adaptation of vision-language models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 25103–25112, 2025. 2

2025
[23]

Test-time adaptation with clip reward for zero-shot general- ization in vision-language models

Shuai Zhao, Xiaohan Wang, Linchao Zhu, and Yi Yang. Test-time adaptation with clip reward for zero-shot general- ization in vision-language models. InInternational Confer- ence on Learning Representations, pages 3597–3613, 2024. 1, 2, 5

2024
[24]

There is{abnormality}

Lihua Zhou, Mao Ye, Shuaifeng Li, Nianxin Li, Xiatian Zhu, Lei Deng, Hongbin Liu, and Zhen Lei. Bayesian test-time adaptation for vision-language models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 29999–30009, 2025. 2 A. Implementation details Adapted parameters.CARVE adapts only the affine scale and ...

2025

[1] [1]

Shah, Greg Zaharchuk, Marc Willis, Adam Yala, Andrew Johnston, Robert D

Louis Blankemeier, Ashwin Kumar, Joseph Paul Cohen, Jiaming Liu, Longchao Liu, Dave Van Veen, Syed Ja- mal Safdar Gardezi, Hongkun Yu, Magdalini Paschali, Zhi- hong Chen, Jean-Benoit Delbrouck, Eduardo Reis, Rob- bie Holland, Cesar Truyts, Christian Bluethgen, Yufu Wu, Long Lian, Malte Engmann Kjeldskov Jensen, Sophie Ost- meier, Maya Varma, Jeya Maria Jo...

2026

[2] [2]

Gradient align- ment improves test-time adaptation for medical image seg- mentation

Zhe Chen, Yu Ye, Yulin Pan, and Yong Xia. Gradient align- ment improves test-time adaptation for medical image seg- mentation. InProceedings of the AAAI Conference on Arti- ficial Intelligence, pages 2429–2437, 2025. 1, 2

2025

[3] [3]

Diverse data augmentation with diffusions for effective test-time prompt tuning

Chun-Mei Feng, Kai Yu, Yong Liu, Salman Khan, and Wangmeng Zuo. Diverse data augmentation with diffusions for effective test-time prompt tuning. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2023. 2

2023

[4] [4]

Domain-adversarial train- ing of neural networks.Journal of Machine Learning Re- search, 17(59):1–35, 2016

Yaroslav Ganin, Evgeniya Ustinova, Hana Ajakan, Pas- cal Germain, Hugo Larochelle, Franc ¸ois Laviolette, Mario Marchand, and Victor Lempitsky. Domain-adversarial train- ing of neural networks.Journal of Machine Learning Re- search, 17(59):1–35, 2016. 1

2016

[5] [5]

Generalist foundation models from a multimodal dataset for 3d computed tomography.arXiv preprint arXiv:2403.17834,

Ibrahim Ethem Hamamci, Sezgin Er, Chenyu Wang, Furkan Almas, Ayse Gulnihan Simsek, Sevval Nil Esir- gun, Irem Dogan, Omer Faruk Durugol, Benjamin Hou, Suprosanna Shit, Weicheng Dai, Murong Xu, Hadrien Rey- naud, Muhammed Furkan Dasdelen, Bastian Wittmann, Tamaz Amiranashvili, Enis Simsar, Mehmet Simsar, Em- ine Bensu Erdemir, Abdullah Alanbay, Anjany Seku...

arXiv

[6] [6]

Efficient test-time adaptation of vision-language models

Adilbek Karmanov, Dayan Guan, Shijian Lu, Abdulmotaleb El Saddik, and Eric Xing. Efficient test-time adaptation of vision-language models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024. 2

2024

[7] [7]

Do we really need to access the source data? source hypothesis transfer for un- supervised domain adaptation

Jian Liang, Dapeng Hu, and Jiashi Feng. Do we really need to access the source data? source hypothesis transfer for un- supervised domain adaptation. InProceedings of the 37th In- ternational Conference on Machine Learning, pages 6028– 6039, 2020. 1

2020

[8] [8]

Ct-glip: 3d grounded language-image pretraining with ct scans and ra- diology reports for full-body scenarios.arXiv preprint arXiv:2404.15272, 2024

Jingyang Lin, Yingda Xia, Jianpeng Zhang, Ke Yan, Kai Cao, Le Lu, Jiebo Luo, and Ling Zhang. Ct-glip: 3d grounded language-image pretraining with ct scans and ra- diology reports for full-body scenarios.arXiv preprint arXiv:2404.15272, 2024. 2

arXiv 2024

[9] [9]

Test-time adaptation with calibration of medical image classification nets for label distribution shift

Xu Ma et al. Test-time adaptation with calibration of medical image classification nets for label distribution shift. InMed- ical Image Computing and Computer Assisted Intervention (MICCAI), 2022. 2

2022

[10] [10]

Efficient test-time model adaptation without forgetting

Shuaicheng Niu, Jiaxiang Wu, Yifan Zhang, Yaofo Chen, Shijian Zheng, Peilin Zhao, and Mingkui Tan. Efficient test-time model adaptation without forgetting. InInterna- tional conference on machine learning, pages 16888–16905. PMLR, 2022. 2

2022

[11] [11]

Towards stable test-time adaptation in dynamic wild world.arXiv preprint arXiv:2302.12400, 2023

Shuaicheng Niu, Jiaxiang Wu, Yifan Zhang, Zhiquan Wen, Yaofo Chen, Peilin Zhao, and Mingkui Tan. Towards stable test-time adaptation in dynamic wild world.arXiv preprint arXiv:2302.12400, 2023. 2

Pith/arXiv arXiv 2023

[12] [12]

Learning transferable visual models from natural language supervision

Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. Learning transferable visual models from natural language supervision. InProceedings of the 38th International Conference on Machine Learning, pages 8748–8763, 2021. 1

2021

[13] [13]

Auto- matic correction of performance drift under acquisition shift in medical image classification.Nature Communications, 14 (1):6608, 2023

M ´elanie Roschewitz, Galvin Khara, Joe Yearsley, Nisha Sharma, Jonathan J James, ´Eva Ambr ´ozay, Adam Heroux, Peter Kecskemethy, Tobias Rijken, and Ben Glocker. Auto- matic correction of performance drift under acquisition shift in medical image classification.Nature Communications, 14 (1):6608, 2023. 1

2023

[14] [14]

Test- time prompt tuning for zero-shot generalization in vision- language models

Manli Shu, Weili Nie, De-An Huang, Zhiding Yu, Tom Goldstein, Anima Anandkumar, and Chaowei Xiao. Test- time prompt tuning for zero-shot generalization in vision- language models. InAdvances in Neural Information Pro- cessing Systems, 2022. 1, 2

2022

[15] [15]

Large-scale and fine-grained vision-language pre-training for enhanced ct image under- standing.arXiv preprint arXiv:2501.14548, 2025

Zhongyi Shui, Jianpeng Zhang, Weiwei Cao, Sinuo Wang, Ruizhe Guo, Le Lu, Lin Yang, Xianghua Ye, Tingbo Liang, Qi Zhang, and Ling Zhang. Large-scale and fine-grained vision-language pre-training for enhanced ct image under- standing.arXiv preprint arXiv:2501.14548, 2025. 1, 2, 5

Pith/arXiv arXiv 2025

[16] [16]

Efros, and Moritz Hardt

Yu Sun, Xiaolong Wang, Zhuang Liu, John Miller, Alexei A. Efros, and Moritz Hardt. Test-time training with self- supervision for generalization under distribution shifts. In Proceedings of the 37th International Conference on Ma- chine Learning, pages 9229–9248, 2020. 1

2020

[17] [17]

Tent: Fully test-time adaptation by entropy minimization

Dequan Wang, Evan Shelhamer, Shaoteng Liu, Bruno Ol- shausen, and Trevor Darrell. Tent: Fully test-time adaptation by entropy minimization. InInternational Conference on Learning Representations (ICLR), 2021. 1, 2, 3, 5, 11

2021

[18] [18]

Continual test-time domain adaptation

Qin Wang, Olga Fink, Luc Van Gool, and Dengxin Dai. Continual test-time domain adaptation. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 7201–7211, 2022. 2

2022

[19] [19]

Multi-label test-time adaptation with bound en- tropy minimization

Xiangyu Wu, Feng Yu, Yang Yang, Qing-Guo Chen, and Jianfeng Lu. Multi-label test-time adaptation with bound en- tropy minimization. InInternational Conference on Learn- ing Representations (ICLR), 2025. 1, 2, 3, 4, 5

2025

[20] [20]

External vali- dation of deep learning algorithms for radiologic diagnosis: a systematic review.Radiology: Artificial Intelligence, 4(3): e210064, 2022

Alice C Yu, Bahram Mohajer, and John Eng. External vali- dation of deep learning algorithms for radiologic diagnosis: a systematic review.Radiology: Artificial Intelligence, 4(3): e210064, 2022. 1

2022

[21] [21]

A large- scale benchmark for test-time adaptation methods in medi- cal image segmentation.arXiv preprint arXiv:2512.02497,

Wenjing Yu, Shuo Jiang, Yifei Chen, Shuo Chang, Yuanhan Wang, Beining Wu, Jie Dong, Mingxuan Liu, Shenghao Zhu, Feiwei Qin, Changmiao Wang, and Qiyuan Tian. A large- scale benchmark for test-time adaptation methods in medi- cal image segmentation.arXiv preprint arXiv:2512.02497,

arXiv

[22] [22]

Realistic test-time adaptation of vision-language models

Maxime Zanella, Cl ´ement Fuchs, Christophe De Vleeschouwer, and Ismail Ben Ayed. Realistic test-time adaptation of vision-language models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 25103–25112, 2025. 2

2025

[23] [23]

Test-time adaptation with clip reward for zero-shot general- ization in vision-language models

Shuai Zhao, Xiaohan Wang, Linchao Zhu, and Yi Yang. Test-time adaptation with clip reward for zero-shot general- ization in vision-language models. InInternational Confer- ence on Learning Representations, pages 3597–3613, 2024. 1, 2, 5

2024

[24] [24]

There is{abnormality}

Lihua Zhou, Mao Ye, Shuaifeng Li, Nianxin Li, Xiatian Zhu, Lei Deng, Hongbin Liu, and Zhen Lei. Bayesian test-time adaptation for vision-language models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 29999–30009, 2025. 2 A. Implementation details Adapted parameters.CARVE adapts only the affine scale and ...

2025