Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

MedMoE: Modality-Specialized Mixture of Experts for Medical Vision-Language Understanding

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Report-conditioned MoE routing improves medical image-text alignment.

desk verdict A useful architectural idea undermined by an experimental design that never isolates the MoE module. read the letter →

arxiv 2506.08356 v2 pith:UHZY3X2N submitted 2025-06-10 cs.CV

Shivang Chopra , Gabriela Sanchez-Rodriguez , Lingchao Mao , Andrew J Feola , Jing Li , Zsolt Kira This is my paper · ORCID
classification cs.CV
keywords Medicalvision-languagemodelsMixture-of-ExpertsModalityspecializationGlobal-localcontrastivelearningZero-shotclassificationimageretrievalSwinTransformerMulti-scalefeatures
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

MedMoE asks whether medical vision-language models should treat every imaging modality the same way when aligning images with reports. The paper's claim is that they should not: a mixture-of-experts module, routed by the diagnostic report, lets the model specialize its visual attention to the granularity of each modality—local lesions for MRI and CT, global patterns for X-ray and ultrasound. The authors build this into a global-local contrastive learning pipeline, train on the UniMed dataset, and report gains in zero-shot classification and linear probing across nine benchmarks spanning four modalities. If the claim holds, modality-specialized routing is a cheap and effective way to improve medical image-text alignment without extra modality labels at inference.

What carries the argument

The central mechanism is the report-conditioned Mixture-of-Experts module over multi-scale Swin features. Each of K convolutional experts takes the feature pyramid $\{F^{(1)},\dots,F^{(L)}\}$, projects every scale to a shared space, and fuses them with per-location soft attention weights $\beta_k^{(\ell)}$; a lightweight MLP maps the global report embedding $t_g$ to a softmax over experts, and hard routing (argmax) selects one expert at inference. The selected local feature grid is then aligned with word-level text embeddings via a GLoRIA-style local contrastive loss, while a cross-entropy auxiliary head on the report embedding supervises expert specialization. This two-level specialization—across scales and across experts—is what carries the argument that modality-adaptive local grounding improves medical vision-language alignment.

What would settle it

Train MedMoE with the MoE module ablated—replacing the expert branches with a single shared head that averages the same multi-scale features—while keeping the Swin Transformer, UniMed data, and all losses identical; if zero-shot accuracy on RSNA, ACL, and CT benchmarks does not drop by a substantial margin, the central claim that routing specializes representations is not supported.

Watch

Extended reading notes

Core claim

MedMoE's central claim is that inserting a report-conditioned Mixture-of-Experts module into the local visual feature extractor—rather than at the fusion or output stage—yields better image-text alignment than a static, modality-agnostic encoder. Each expert is a small convolutional head that fuses Swin Transformer feature pyramids via cross-scale attention, and a hard router selects exactly one expert per input based on the global report embedding. An auxiliary loss over the report type nudges experts toward diagnostic modalities, so at inference the model needs only the report text, not a modality label, to route. The paper reports that this design improves or matches generalist and specialist baselines on most of nine zero-shot benchmarks and several linear-probing settings, with attention visualizations showing experts concentrating on modality-appropriate anatomy.

Load-bearing premise

The central claim is that the reported accuracy gains come from the report-conditioned MoE routing, but no experiment removes the MoE while keeping the same Swin backbone and the same UniMed pretraining data, so the gains could in principle come from the backbone or the data.

Editorial extensions

If this is right

  • Zero-shot classification improves on 6 of 9 radiology benchmarks, including CT views where prior generalist models were weakest.
  • Hard routing keeps inference FLOPs near a single Swin-Tiny branch (7.8 GFLOPs, ~37M parameters) despite training many experts, so specialization does not cost runtime.
  • The router needs only the report text at inference, not a modality label, meaning the model can specialize without extra annotation.
  • Freezing MedMoE's encoder and fitting a linear head transfers well in low-data regimes, beating prior state of the art on RSNA at 1% training data.
  • Attention maps show each expert attends to modality-relevant anatomy, supporting the claim that routing changes spatial grounding rather than just scale.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A direct ablation that disables the MoE while keeping the same Swin backbone and UniMed pretraining data would make the mechanism's contribution measurable; the paper's tables compare against different backbones and datasets.
  • Report-conditioned routing could be extended to new modalities such as fundus or pathology by appending new expert branches and training only those branches, keeping inference cost flat.
  • Because only one expert is active per input, MedMoE naturally supports adding modalities at runtime without scaling inference, and may enable continual learning by adding experts for new modalities.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes MedMoE, a medical vision-language model that augments a Swin Transformer backbone with a report-conditioned Mixture-of-Experts (MoE) module. The router selects one of several convolutional expert branches based on the global report embedding, and the selected expert fuses multi-scale image features through cross-scale attention. Training uses global and local contrastive losses plus an auxiliary report-type classification loss. The authors report zero-shot classification and linear probing results across X-ray, ultrasound, MRI, and CT benchmarks, claiming improvements over generalist baselines such as UniMed-CLIP and PMC-CLIP, along with qualitative attention visualizations and a computational cost comparison.

Significance. If the central claim were fully supported, the idea of modality-specialized expert routing inside the visual encoder would be a useful contribution to medical vision-language pretraining, where static encoders are the norm. The paper uses publicly available data and benchmarks, compares against several strong baselines, and provides a parameter/FLOP analysis showing that hard routing keeps inference cost close to a single Swin-T backbone. However, the experimental design does not currently isolate the MoE mechanism, and several text/table contradictions undermine confidence in the reported numbers. The contribution is therefore promising but not demonstrated in the present manuscript.

major comments (4)
  1. [§5.2, Table 1] The central claim that report-conditioned MoE routing improves performance is not supported by any ablation that isolates the MoE module. MedMoE differs from UniMed-CLIP in backbone (Swin-T vs ViT-B/16), initialization (MedCLIP-pretrained weights), and training objective (addition of the auxiliary report-type loss). A K=1 baseline without a router, trained on the same Swin-T backbone, the same UniMed 5.3M data, and the same global/local/auxiliary losses, is needed to attribute the reported gains to the MoE mechanism. Without it, the differences in Table 1 (e.g., +6.67 on RSNA, -10.40 on Meniscus relative to UniMed-CLIP) could be caused by any of these changes.
  2. [§4.3, §4.5, Abstract] The router is conditioned on the global report embedding t_g and trained with an auxiliary loss using the ground-truth report type y, but zero-shot classification (Table 1) is performed without a report. The paper never states what t_g is during zero-shot inference. If the class prompts encode modality names (e.g., 'a CT image of ...'), then the router is receiving modality information at test time, contradicting the abstract's claim that the method works 'without requiring modality-specific supervision at inference.' This ambiguity is load-bearing because it determines whether the zero-shot comparison is a fair test of the proposed mechanism.
  3. [§5.2, Tables 1 and 2] Several empirical claims contradict the reported tables. Section 5.2 states that MedMoE delivers state-of-the-art results on all three CT views, but Table 1 shows UniMed-CLIP achieving 31.72 on sagittal CT versus MedMoE's 26.83. The same paragraph says MedMoE attains the best performance on 6/9 datasets, whereas Table 1 shows best results on 5/9 (CheXpert, RSNA, Breast, axial CT, coronal CT). Section 5.3 also claims that for MRI (ACL) and CT (MediMeTA Axial) MedMoE matches or exceeds all generalist baselines, but Table 2 shows UniMed-CLIP reaching 97.28 on ACL 100% versus MedMoE's 92.84, and a tie on MediMeTA Axial 100% at 76.38. These inconsistencies need to be corrected or explained.
  4. [Tables 1 and 2] No error bars, standard deviations, or significance tests are reported for any experiment. Many comparisons in Table 1 are small (e.g., 66.03 vs 65.90 on CheXpert), and without repeated runs it is impossible to assess whether the reported gains are meaningful. At minimum, the authors should report variance over multiple seeds or a statistical test for the main zero-shot and linear probing comparisons.
minor comments (5)
  1. [§3/§4] The section numbering is duplicated: '3. Method' is immediately followed by '4. Methodology,' and the latter contains the actual method description. The duplicate header should be removed and sections renumbered.
  2. [Figure 2 and §4.3] The Figure 2 caption states that the router selects an expert 'based on the global image embedding,' while §4.3 defines the router as conditioned on the global report embedding t_g. These are different inputs; the inconsistency should be resolved.
  3. [§5.1, References] Reference [4] is cited for 'MedCLIP-pretrained weights,' but the reference is the ImageNet paper by Deng et al. The citation is incorrect or the provenance of the initialization weights needs a proper source.
  4. [§5.2] The sentence 'outperforming the strongest generalist baseline (PMC-CLIP) by 3.33% and 0.56%, respectively' is unclear: PMC-CLIP is not the strongest generalist baseline in Table 1 (UniMed-CLIP has higher average accuracy), and the two percentages are not tied to specific datasets.
  5. [Table 1] The column header 'Avg.CheXpert(5x200)' appears to be a formatting error, merging the average column with the CheXpert column. The table would be easier to read with separate 'Avg.' and 'CheXpert' headers.

Circularity Check

1 steps flagged · score 5.0 of 10

The claimed modality specialization of MedMoE's experts is directly supervised by ground-truth modality labels (Laux), making that specialization a construction rather than an emergent result; the retrieval/classification numbers are not derived from the label, but they are confounded by the absence of a matched no-MoE baseline.

  1. fitted input called prediction [Section 4.5 (Auxiliary Report-Type Supervision), Eq. (5); see also Section 4.3 and Figure 2.]
    "Laux = CrossEntropy(W_c · t_g, y), where W_c is a classification head and y is the ground-truth report type (e.g., CT, X-ray, MRI). This encourages the router to learn modality-sensitive routing that aligns with variations in real diagnostic contexts."

    The paper's central claim that the MoE experts are modality-specialized and 'capture modality-specific visual semantics' (Abstract) is produced by this explicit loss: the router α = softmax(W_2·ReLU(W_1·t_g)) is supervised with the ground-truth modality label y, and the experts are then labeled CT/MRI/X-ray/Ultrasound in Figure 2 and in the text. The observed specialization is therefore a direct, by-construction consequence of the fitted auxiliary classifier, not an emergent property of report-conditioned routing. The downstream zero-shot and linear-probing accuracies are not derived from y, so the circularity is partial rather than total.

full rationale

No load-bearing self-citation chain appears; the cited works (GLoRIA, UniMed-CLIP, MedCLIP, etc.) are external and are not used to justify the key mechanism. The MoE router's modality specialization, however, is explicitly trained with ground-truth modality labels through Laux, and the paper then presents that specialization as a learned result ('modality-specialized expert branches', 'each expert ... focus on distinct diagnostic modalities'). This is the main circular step: the 'discovery' that experts specialize by modality is built into the training objective by construction. The reported accuracy and retrieval numbers in Tables 1-2 are not logically forced by the modality labels, so the empirical contribution retains independent content. The missing no-MoE control (same Swin-T backbone, same UniMed data, same losses, with K=1 and no router) is a serious experimental confound but not, by itself, a circularity. Because the specialization claim reduces to a supervised fit while the main downstream numbers do not, a moderate score of 5 is appropriate.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The method is an empirical architecture with several hand-chosen components. The main claim depends on an untested baseline attribution and on assumptions about the text-based router.

free parameters (4)
  • auxiliary loss weight λ = not reported
    The balance between contrastive and auxiliary losses is not given; chosen by hand without sensitivity analysis.
  • temperature τ = not reported
    Used in global and local contrastive losses; value not reported.
  • number of experts K = 4
    Four experts (CT, MRI, X-ray, Ultrasound); no ablation on the number of experts.
  • expert branch architecture = 3-layer CNN with BatchNorm and ReLU
    No comparison to other expert capacities or depths.
assumptions (4)
  • domain assumption UniMed dataset contains sufficient paired image-text data with reliable modality labels for pretraining and auxiliary supervision.
    Section 5.1 says UniMed is used for pretraining and the auxiliary loss requires ground-truth report types.
  • domain assumption The global report embedding contains enough modality information for routing.
    Section 4.3 conditions the router solely on the text embedding t_g.
  • domain assumption MedCLIP-pretrained Swin weights provide a suitable initialization.
    Section 5.1 implementation details initialize from MedCLIP-pretrained weights [4].
  • ad hoc to paper Hard routing with argmax is trainable via the auxiliary loss only, and this is sufficient for expert specialization.
    The paper does not describe any gradient approximation for the hard selection, so the router appears to receive no signal from the local contrastive loss.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MedMoE: Modality-Specialized Mixture of Experts for Medical Vision-Language Understanding." pith.science (2026). https://pith.science/paper/UHZY3X2N

@misc{pith2026250608356,
  author       = {Pith},
  title        = {Pith review of: MedMoE: Modality-Specialized Mixture of Experts for Medical Vision-Language Understanding},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UHZY3X2N}},
  note         = {Machine review of arXiv:2506.08356}
}
read the original abstract

Different medical imaging modalities capture diagnostic information at varying spatial resolutions, from coarse global patterns to fine-grained localized structures. However, most existing vision-language frameworks in the medical domain apply a uniform strategy for local feature extraction, overlooking the modality-specific demands. In this work, we present MedMoE, a modular and extensible vision-language processing framework that dynamically adapts visual representation based on the diagnostic context. MedMoE incorporates a Mixture-of-Experts (MoE) module conditioned on the report type, which routes multi-scale image features through specialized expert branches trained to capture modality-specific visual semantics. These experts operate over feature pyramids derived from a Swin Transformer backbone, enabling spatially adaptive attention to clinically relevant regions. This framework produces localized visual representations aligned with textual descriptions, without requiring modality-specific supervision at inference. Empirical results on diverse medical benchmarks demonstrate that MedMoE improves alignment and retrieval performance across imaging modalities, underscoring the value of modality-specialized visual representations in clinical vision-language systems.

Figures

Figures reproduced from arXiv: 2506.08356 by the authors.

Figure 1
Figure 1. Motivation: diagnostic context varies across modali￾ties. Each image-report pair exemplifies how clinical observations may focus on localized lesions (red) or broader regions (blue), de￾pending on the imaging modality and diagnostic task. fine-grained semantic alignment between regions in an im￾age and words in a report [6]. However, current approaches typically employ a static strategy for feature extraction, apply… view at source ↗
Figure 2
Figure 2. Overall architecture of MedMoE. A Swin Transformer extracts multi-scale features, and a modality-aware router selects a specialized expert based on the global image embedding. The expert outputs local embeddings aligned with word-level text via a local contrastive loss. Global contrastive loss and auxiliary loss enforce image-text alignment and supervise expert selection respectively. where each F (l) corresponds to… view at source ↗
Figure 3
Figure 3. Visualization of Attention Weights across Modalities. Word-level attention maps are shown for different imaging modalities (X-ray, Ultrasound, MRI, and CT), highlighting regions relevant to the associated diagnostic labels. ( [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. LocAnyMed: Vision-Language Grounding for Multimodal Medical Images

    cs.CV 2026-08 conditional novelty 7.0 of 10

    A unified 210K medical grounding dataset and full fine-tuning raise a general model's localization F1 from 10.6 to 85.6 on held-out data and improve cross-source transfer.

Reference graph

Works this paper leans on

28 extracted references · 20 canonical work pages · cited by 1 Pith paper

  1. [1]

    Prior: Prototype-driven radiograph interpretation with optimized representations

    Bowen Cheng and et al. Prior: Prototype-driven radiograph interpretation with optimized representations. InNeurIPS,

    work page

  2. [2]

    The rsna pulmonary embolism ct dataset.Radiology: Artificial Intelligence, 3(2):e200254, 2021

    Errol Colak, Felipe C Kitamura, Stephen B Hobbs, Carol C Wu, Matthew P Lungren, Luciano M Prevedello, Jayashree Kalpathy-Cramer, Robyn L Ball, George Shih, Anouk Stein, et al. The rsna pulmonary embolism ct dataset.Radiology: Artificial Intelligence, 3(2):e200254, 2021. 4

    work page 2021

  3. [3]

    Damai Dai, Chengqi Deng, Chenggang Zhao, R. X. Xu, Huazuo Gao, Deli Chen, Jiashi Li, Wangding Zeng, Xingkai Yu, Y . Wu, Zhenda Xie, Y . K. Li, Panpan Huang, Fuli Luo, Chong Ruan, Zhifang Sui, and Wenfeng Liang. Deepseek- moe: Towards ultimate expert specialization in mixture-of- experts language models, 2024. 2

    2024

  4. [4]

    Imagenet: A large-scale hierarchical image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In2009 IEEE Conference on Computer Vision and Pattern Recognition, pages 248–255. IEEE, 2009. 4

    work page 2009

  5. [5]

    Yuxin et al. Du. Glam: Efficient scaling of language models with mixture-of-experts. InICML, 2022. 2

    work page 2022

  6. [6]

    Lungren, and Serena Yeung

    Shih-Cheng Huang, Liyue Shen, Matthew P. Lungren, and Serena Yeung. Gloria: A multimodal global-local represen- tation learning framework for label-efficient medical image recognition. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2021. 1, 2, 3, 4

    work page 2021

  7. [7]

    Quilt-1m: One million image-text pairs for histopathology.Advances in neural information processing systems, 36:37995–38017,

    Wisdom Ikezogwo, Saygin Seyfioglu, Fatemeh Ghezloo, Dylan Geva, Fatwir Sheikh Mohammed, Pavan Kumar Anand, Ranjay Krishna, and Linda Shapiro. Quilt-1m: One million image-text pairs for histopathology.Advances in neural information processing systems, 36:37995–38017,

    work page

  8. [8]

    Chexpert: A large chest radiograph dataset with uncertainty labels and expert comparison

    Jeremy Irvin, Pranav Rajpurkar, Michael Ko, Yifan Yu, Sil- viana Ciurea-Ilcus, Chris Chute, Henrik Marklund, Behzad Haghgoo, Robyn Ball, Katie Shpanskaya, et al. Chexpert: A large chest radiograph dataset with uncertainty labels and expert comparison. InProceedings of the AAAI Conference on Artificial Intelligence, pages 590–597, 2019. 4

    work page 2019

Show all 28 references
  1. [9]

    Unimed- clip: Towards a unified image-text pretraining paradigm for diverse medical imaging modalities.arXiv preprint arXiv:2412.10372, 2024

    Muhammad Uzair Khattak, Shahina Kunhimon, Muzammal Naseer, Salman Khan, and Fahad Shahbaz Khan. Unimed- clip: Towards a unified image-text pretraining paradigm for diverse medical imaging modalities.arXiv preprint arXiv:2412.10372, 2024. 1, 4

    2024 arXiv

  2. [10]

    Lepikhin

    Dmitry et al. Lepikhin. Gshard: Scaling giant models with conditional computation and automatic sharding. InICLR,

  3. [11]

    Moe-llava: Mixture of experts for large vision-language models.arXiv preprint arXiv:2401.15947, 2024

    Bin Lin, Zhenyu Tang, Yang Ye, Jiaxi Cui, Bin Zhu, Peng Jin, Junwu Zhang, Munan Ning, and Li Yuan. Moe-llava: Mixture of experts for large vision-language models.arXiv preprint arXiv:2401.15947, 2024. 2

    2024 arXiv

  4. [12]

    Pmc-clip: Con- trastive language-image pre-training using biomedical docu- ments

    Weixiong Lin, Ziheng Zhao, Xiaoman Zhang, Chaoyi Wu, Ya Zhang, Yanfeng Wang, and Weidi Xie. Pmc-clip: Con- trastive language-image pre-training using biomedical docu- ments. InInternational Conference on Medical Image Com- puting and Computer-Assisted Intervention, pages 525–5...

    2023

  5. [13]

    Visual instruction tuning, 2023

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning, 2023. 1

    2023

  6. [14]

    Swin transformer: Hierarchical vision transformer using shifted windows

    Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin transformer: Hierarchical vision transformer using shifted windows. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2021. 2, 4

    2021

  7. [15]

    Lovt: Local vision-text alignment for medical image analysis

    Jonas Mueller and et al. Lovt: Local vision-text alignment for medical image analysis. InMICCAI, 2022. 2

    2022

  8. [16]

    Quiltnet: Efficient deep learning inference on multi-chip accelerators using model partitioning

    Jongho Park, HyukJun Kwon, Seowoo Kim, Junyoung Lee, Minho Ha, Euicheol Lim, Mohsen Imani, and Yeseong Kim. Quiltnet: Efficient deep learning inference on multi-chip accelerators using model partitioning. InProceedings of the 59th ACM/IEEE Design Automation Conference, pages 1...

    2022

  9. [17]

    An open access thy- roid ultrasound image database

    Lina Pedraza, Carlos Vargas, Fabi ´an Narv´aez, Oscar Dur´an, Emma Mu ˜noz, and Eduardo Romero. An open access thy- roid ultrasound image database. In10th International sympo- sium on medical information processing and analysis, pages 188–193. SPIE, 2015. 4

    2015

  10. [18]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Jack Hallacy, Aditya Ramesh, et al. Learning transferable visual models from natural language supervision. InProceedings of the Inter- national Conference on Machine Learning (ICML), 2021. 1, 4

    2021

  11. [19]

    Noam et al. Shazeer. Outrageously large neural networks: The sparsely-gated mixture-of-experts layer. InICLR, 2017. 2

    2017

  12. [20]

    Yunhao et al. Shui. fvlm: Anatomy-aware vision-language model for volumetric medical imaging.arXiv preprint arXiv:2503.00000, 2025. 2

    2025

  13. [21]

    Large language models encode clinical knowledge.Nature, 620 (7972):172–180, 2023

    Karan Singhal, Shekoofeh Azizi, Tao Tu, S Sara Mahdavi, Jason Wei, Hyung Won Chung, Nathan Scales, Ajay Tan- wani, Heather Cole-Lewis, Stephen Pfohl, et al. Large language models encode clinical knowledge.Nature, 620 (7972):172–180, 2023. 1

    2023

  14. [22]

    Medclip: Contrastive learning from unpaired medical images and text

    Zifeng Wang, Zhenbang Wu, Dinesh Agarwal, and Jimeng Sun. Medclip: Contrastive learning from unpaired medical images and text. InProceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 3876–3887, Abu Dhabi, United Arab Emirates, 2022. Asso...

    2022

  15. [23]

    Mm-retinal: Knowledge-enhanced founda- tional pretraining with fundus image-text expertise

    Ruiqi Wu, Chenran Zhang, Jianle Zhang, Yi Zhou, Tao Zhou, and Huazhu Fu. Mm-retinal: Knowledge-enhanced founda- tional pretraining with fundus image-text expertise. InIn- ternational Conference on Medical Image Computing and Computer-Assisted Intervention, pages 722–732. Springer,

  16. [24]

    Deepseek-vl2: Mixture-of-experts vision-language models for advanced multimodal understanding, 2024

    Zhiyu Wu, Xiaokang Chen, Zizheng Pan, Xingchao Liu, Wen Liu, Damai Dai, Huazuo Gao, Yiyang Ma, Chengyue Wu, Bingxuan Wang, Zhenda Xie, Yu Wu, Kai Hu, Jiawei Wang, Yaofeng Sun, Yukun Li, Yishi Piao, Kang Guan, Aixin Liu, Xin Xie, Yuxiang You, Kai Dong, Xingkai Yu, Haowei Zhang,...

    2024

  17. [25]

    Biomedclip: A mul- timodal biomedical foundation model pretrained from fif- teen million scientific image-text pairs.arXiv preprint arXiv:2303.00915, 2023

    Sheng Zhang, Yanbo Xu, Naoto Usuyama, Hanwen Xu, Jaspreet Bagga, Robert Tinn, et al. Biomedclip: A mul- timodal biomedical foundation model pretrained from fif- teen million scientific image-text pairs.arXiv preprint arXiv:2303.00915, 2023. 4

    2023 arXiv

  18. [26]

    Clip in medical imaging: A survey.arXiv preprint arXiv:2312.07353, 2025

    Zihao Zhao, Yuxiao Liu, Han Wu, et al. Clip in medical imaging: A survey.arXiv preprint arXiv:2312.07353, 2025. 1, 2

    2025 arXiv

  19. [27]

    Zhao, Andrew Dai, Zhifeng Chen, Quoc Le, and James Laudon

    Yanqi Zhou, Tao Lei, Hanxiao Liu, Nan Du, Yanping Huang, Vincent Y . Zhao, Andrew Dai, Zhifeng Chen, Quoc Le, and James Laudon. Mixture-of-experts with expert choice rout- ing. InAdvances in Neural Information Processing Systems (NeurIPS), 2022. 2

    2022

  20. [28]

    Advancing radiograph representation learning with masked record mod- eling.ICLR, 2023

    Zongyu Zhou, Yuxuan Zhou, Yifan Wang, et al. Advancing radiograph representation learning with masked record mod- eling.ICLR, 2023. 1, 2

    2023

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.