Pith. sign in

REVIEW 4 major objections 5 minor 3 cited by

Hi-End-MAE: Hierarchical encoder-driven masked autoencoders are stronger vision learners for medical image segmentation

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

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

Pith's one-line read Hi-End-MAE claims that shifting reconstruction from the decoder to the encoder in masked image modeling produces stronger medical image representations, reporting a 69.25% average one-shot Dice across six CT datasets.

desk verdict Solid method and ablations, but the 6% headline gap is mostly model capacity; the same-backbone gain is only ~1–2 points, so the claim needs apples-to-apples reporting. read the letter →

arxiv 2502.08347 v1 pith:NGC5PE4Y submitted 2025-02-12 cs.CV

Fenghe Tang , Qingsong Yao , Wenxin Ma , Chenxu Wu , Zihang Jiang , S. Kevin Zhou This is my paper · ORCID
classification cs.CV
keywords maskedimagemodelingencoder-drivenreconstructionhierarchicaldensedecodingmedicalpre-trainingsegmentationVisionTransformerself-supervisedlearningone-shot
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

This paper claims that masked autoencoders for medical images should shift the burden of reconstruction from the decoder to the encoder. In standard MAE, the decoder rebuilds masked patches from encoder outputs, so the encoder is only indirectly responsible for reconstruction quality; Hi-End-MAE instead has decoding tokens query the encoder's own visible-patch representations, and repeats this at several encoder depths, forcing the encoder to store the anatomical detail needed to recover what was masked. The authors pre-train on roughly 10,000 CT scans and report that the resulting encoder transfers better to downstream organ segmentation, reaching a 69.25% average Dice score (a standard overlap metric) with a single annotated volume across six CT datasets, at least 6.04 points above all compared methods, while cutting decoder computation by about 35%. If this holds, it argues that the encoder, not the decoder, should be the primary learning target in masked image modeling for label-scarce medical imaging.

What carries the argument

The load-bearing mechanism is the encoder-driven cross-attention decoder. Decoding tokens, after a self-attention step, query only the visible encoder tokens as keys and values, so reconstruction quality is tied directly to encoder representation quality; the hierarchy repeats this at encoder layers 3, 6, and 9 in a bottom-up scheme, mirroring UNETR's skip connections for upstream-downstream alignment. This changes the attention complexity from $\mathcal{O}(N^2 d_k)$ for a full-token decoder to $\mathcal{O}(N M d_k) = \mathcal{O}(N^2(1-\gamma)d_k)$ with mask ratio $\gamma = 0.75$, which is the formal source of the reported ~35% decoder savings. The paper also uses effective rank and singular-value distributions of the attention values $V$ as its quantitative evidence that the encoder learns richer features.

What would settle it

Pre-train Hi-End-MAE and vanilla MAE with the same encoder size (e.g., both ViT-B/12(1536)) and equal decoder parameter budgets, then fine-tune both on the same six one-shot tasks and compare average Dice: if the gap stays near the 1.13 points of Table 6 rather than the 6.04 points of Table 2, the headline advantage is attributable to model size rather than to the encoder-driven mechanism.

Watch

Extended reading notes

Core claim

The central claim is that decoder-driven reconstruction, the design shared by MAE and its medical variants, underuses the encoder and therefore produces weaker features for medical segmentation. Hi-End-MAE replaces it with encoder-driven reconstruction: decoder tokens first self-attend, then cross-attend to the visible encoder tokens at three ascending layers (3, 6, and 9), so that only visible tokens serve as keys and values and the reconstruction loss directly measures the encoder's representational sufficiency. The paper reports that on one-shot segmentation over six CT datasets Hi-End-MAE reaches 69.25% average Dice, at least 6.04 points above MAE and all medical self-supervised baselines, with a 34.97% reduction in decoder FLOPS; it also reports gains across 1%, 10%, and 100% fine-tuning proportions, cross-modality transfer to MRI (61.45% on BraTS21), monotonic gains as pre-training data grow from 1K to 10K scans, and higher effective rank and singular values of attention maps as evidence of richer representations.

Load-bearing premise

The headline superiority of 69.25% versus 63.21% rests on a comparison in which Hi-End-MAE uses the larger ViT-B/12(1536) encoder while the MAE baseline uses the smaller ViT-B/16(768), so if the larger model size produces most of the gain, the method's actual contribution is the roughly 1.13% average improvement shown in the equal-backbone comparison of Table 6.

Editorial extensions

If this is right

  • A single annotated volume yields 69.25% average Dice across six CT benchmarks, a 32.34-point gain over training from scratch.
  • At 1%, 10%, and 100% data proportions across AMOS, FLARE'22, and BTCV, Hi-End-MAE averages 75.72% Dice, 2.59 points above the strongest medical SSL baseline.
  • Decoder FLOPS drop by about 35% at a 75% mask ratio, and the saving grows linearly as the mask ratio rises because only visible tokens serve as keys and values.
  • A CT-pretrained encoder transfers to MRI: one-shot Dice on BraTS21 is 61.45%, above every compared method.
  • Growing pre-training data from 1K to 10K scans raises one-shot Dice from 63.32% to 69.25% and 10%-proportion Dice from 71.75% to 78.21%.

Reading between the lines

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

  • A matched-backbone experiment, with identical encoder size for Hi-End-MAE and MAE, is the natural next test to separate the mechanism's contribution from the larger ViT-B/12(1536) backbone used in the headline comparison.
  • The decoding hierarchy is aligned to UNETR's skip-connection depths, so whether the pre-trained encoder retains its advantage behind other segmentation heads is an untested consequence of the up-downstream alignment argument.
  • The ablation tables sketch a cost-accuracy frontier: mask ratios above 75% buy extra compute savings but steadily lower one-shot Dice, so a stronger encoder that shifts that frontier would be a direct follow-up.
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 Hi-End-MAE, a self-supervised masked image modeling pre-training method for 3D medical image segmentation. The method replaces the decoder-driven reconstruction of MAE with an encoder-driven reconstruction in which decoder query tokens cross-attend to visible encoder tokens, and it adds hierarchical dense decoding that queries encoder features at layers 3, 6, and 9. The model is pre-trained on 9,995 CT scans and evaluated on one-shot segmentation across six CT datasets plus the BraTS21 MRI dataset, and on 1%, 10%, and 100% fine-tuning on AMOS, FLARE'22, and BTCV. The paper reports a one-shot average DSC of 69.25%, at least 6.04% higher than compared baselines, and a 34.97% reduction in decoder FLOPs relative to MAE.

Significance. The encoder-driven reconstruction and hierarchical dense decoding ideas are well motivated, and the paper includes informative ablations (Table 7) and a controlled same-backbone comparison (Table 6) that supports a modest but consistent improvement over MAE. The release of code is a strength. However, the headline one-shot result is based on a comparison that mixes model size with pre-training objective, so the central 'stronger vision learner' claim is not established by the presented evidence. The efficiency analysis contains a technical error. With corrected framing and additional controlled experiments, the method could be a useful contribution to medical image pre-training.

major comments (4)
  1. [§4.3, Table 2] The headline claim that Hi-End-MAE 'achieves a DSC score of 69.25%, at least 6.04% higher than all compared methods' compares Hi-End-MAE with a ViT-B/12(1536) encoder (per Table 7) against MAE with a ViT-B/16(768) encoder and against SSL baselines using different network architectures (3D U-Net, SwinUNETR, MedNeXt). This is a systems-level comparison, not a comparison of pre-training objectives. The paper's own controlled same-backbone comparison in Table 6 shows only a 1.13% average DSC improvement over MAE (75.05 vs 73.92) on AMOS/FLARE/BTCV, and Table 7 shows a 2.33% one-shot gap (69.25 vs 66.92). The authors should either present the headline as a systems-level result or base the method-specific claim on the same-backbone comparisons; otherwise the 'stronger vision learner' claim is overstated.
  2. [§4.1 (Table 1) and §4.3] Four of the six one-shot downstream datasets (BTCV, CT-ORG, Sliver07, FLARE'22) are included in the 10K CT pre-training corpus. Because the pre-trained baselines use official weights that may have different pre-training corpora, the one-shot numbers include an uncontrolled overlap advantage. The authors should report the one-shot results on the unseen datasets (WORD and AMOS) separately as the primary generalization evidence, and ideally fine-tune all baselines on the same pre-training data to control for data overlap.
  3. [§4.3 and Table 2] One-shot DSC scores are reported as single numbers without error bars or multiple seeds. With only one training volume, run-to-run variance can easily exceed the 2.33% same-backbone gap reported in Table 7; this is particularly important because the central method comparison is a difference of a few points. Reporting mean and standard deviation over at least three seeds would substantiate the one-shot results.
  4. [§3.2, Eq. (7)] The statement 'the first term in Eq. (7) is dominant' is not correct for the reported configuration: with patch size 12 on 96^3 volumes, N ≈ 512, and with d_k = 1536, N·d_k^2 ≈ 1.2e9 exceeds N^2(1−γ)·d_k ≈ 1.0e8 for γ = 0.75. Consequently, the claimed 'linear with mask ratio' savings do not follow from the complexity expression. The empirical 34.97% FLOPs reduction in Table 6 may be plausible, but it should be backed by a corrected complexity accounting.
minor comments (5)
  1. [§4.4, Fig. 6 and Fig. 7] The effective-rank and singular-value analyses are presented as evidence of representation quality, but no quantitative link to downstream segmentation performance is established; these figures should be described as descriptive post-hoc analyses.
  2. [§4.2 and Table 2] The text says 'we adopt UNETR as the downstream networks,' but Table 2 includes rows for SwinUNETR, MedNeXt, and other backbones; the table should clarify which rows use which downstream network and which rows load official weights.
  3. [Table 3] The column header 'Avg' appears in several places (AMOS Avg, FLARE'22 Avg, BTCV Avg, and a final '‡ Avg'), and it is not clear whether the final average is over datasets, over proportions, or both; a footnote defining the averaging procedure would help.
  4. [§3.2, Eq. (6)] There are small presentation errors in the definition of effective rank: 'matirx' should be 'matrix', and the normalized singular value uses an undefined denominator 'P kσi', which should be written as Σ_j σ_j.
  5. [§4.1] The phrase 'interpolated to the isotropic voxel spacing' is better rendered as 'resampled to an isotropic voxel spacing of 1.5 mm'; interpolation is only one part of the resampling operation.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular reasoning found: the method's gains are empirical and its complexity reduction follows from the stated architecture.

full rationale

The paper makes no derivation that reduces to its inputs. The central claims are empirical: pre-training Hi-End-MAE and fine-tuning on downstream segmentation benchmarks. The encoder-driven reconstruction mechanism is an architectural design, and the claim that it 'compels the encoder to learn stronger representations' is a causal interpretation, not a definitional equivalence. The computational complexity reduction (Eq. 7) is a direct consequence of only attending to visible tokens, which is a stated design property rather than a fitted prediction. The representation analyses (effective rank, singular values, attention maps) are post-hoc measurements, not predictions derived from the method. Comparisons in Table 2 mix backbone sizes, which is a legitimate experimental fairness concern, but it does not make the reported numbers circular: the results are external measurements on held-out validation sets. Self-citations to HySparK and Zhou et al. are used for context or comparison, not as load-bearing justification. Therefore no circular step meets the quoted-evidence threshold.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The method's empirical gains rest on standard ML choices: a hand-picked mask ratio, decoder depth, and a specific encoder variant. No new physical entities or mathematical axioms are introduced; the assumptions are dataset and architecture transferability.

free parameters (3)
  • mask_ratio (γ) = 0.75
    Chosen via ablation across 75-95% in Table 5; 75% gave the best one-shot DSC average.
  • decoder_stages (B) = 3
    Chosen via ablation in Table 7; adding stages progressively improved one-shot performance and reduced FLOPS.
  • encoder variant = ViT-B/12(1536)
    Main results use the largest variant tested; Table 7 shows it outperforms smaller variants, but the gain may be due to capacity rather than the method.
assumptions (3)
  • domain assumption Effective rank and singular value distributions of attention values are valid proxies for representation quality.
    Used in Section 4.4 to argue Hi-End-MAE learns richer representations; no causal link to downstream performance is established.
  • domain assumption Pre-training on CT sub-volumes resampled to 1.5mm isotropic spacing transfers to downstream MRI segmentation (BraTS21).
    Section 4.3, Table 4; the paper claims low-level representation transfer across modalities without an explicit mechanism.
  • ad hoc to paper The pre-training decoder stages aligned to UNETR skip-connection layers (3, 6, 9) are a universally good choice.
    Section 4.2 sets decoding stages to match UNETR; no experiments test other alignment strategies, so the design is specific to this downstream architecture.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Hi-End-MAE: Hierarchical encoder-driven masked autoencoders are stronger vision learners for medical image segmentation." pith.science (2026). https://pith.science/paper/NGC5PE4Y

@misc{pith2026250208347,
  author       = {Pith},
  title        = {Pith review of: Hi-End-MAE: Hierarchical encoder-driven masked autoencoders are stronger vision learners for medical image segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NGC5PE4Y}},
  note         = {Machine review of arXiv:2502.08347}
}
read the original abstract

Medical image segmentation remains a formidable challenge due to the label scarcity. Pre-training Vision Transformer (ViT) through masked image modeling (MIM) on large-scale unlabeled medical datasets presents a promising solution, providing both computational efficiency and model generalization for various downstream tasks. However, current ViT-based MIM pre-training frameworks predominantly emphasize local aggregation representations in output layers and fail to exploit the rich representations across different ViT layers that better capture fine-grained semantic information needed for more precise medical downstream tasks. To fill the above gap, we hereby present Hierarchical Encoder-driven MAE (Hi-End-MAE), a simple yet effective ViT-based pre-training solution, which centers on two key innovations: (1) Encoder-driven reconstruction, which encourages the encoder to learn more informative features to guide the reconstruction of masked patches; and (2) Hierarchical dense decoding, which implements a hierarchical decoding structure to capture rich representations across different layers. We pre-train Hi-End-MAE on a large-scale dataset of 10K CT scans and evaluated its performance across seven public medical image segmentation benchmarks. Extensive experiments demonstrate that Hi-End-MAE achieves superior transfer learning capabilities across various downstream tasks, revealing the potential of ViT in medical imaging applications. The code is available at: https://github.com/FengheTan9/Hi-End-MAE

Figures

Figures reproduced from arXiv: 2502.08347 by the authors.

Figure 1
Figure 1. Decoder-driven vs. encoder-driven reconstruction. Conventional MAE is based on (a) decoder-driven reconstruction and Hi-End-MAE is based on (b) [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Performance comparisons against well-known medical self-supervised [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. The overall framework of Hi-End-MAE. The Encoder-driven Dense Decoding architecture uses encoder representations to guide the decoder bottom-up dense reconstruction. The encoder (blue) is a Vision Transformer (ViT), which only processes the visible patches (blue cube). The decoder (green) incorporates a cross-attention mechanism, feeding in a full set of token i.e. visible token (grey cube) and learnable masked toke… view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Comparative analysis one-shot segmentation results across 16 targets in terms of DSC (%) performance. The abbreviations Lun, Liv, Kid, Vein, Aor, Int, Spl, [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Visualization of attention maps in the 3th, 6th, 9th, and 12th layers of ViT-B [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Comparison effective rank (Roy and Vetterli, 2007) distribution of attention values (V) with MAE and Hi-End-MAE in ViT-B/12(1536). A rightward shift indicates a richer effective rank, reflecting more diverse data representations and a more uniform distribution. This en…
Figure 7
Figure 7. Figure 7: Comparison singular values distribution of attention values ( [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: Qualitative visualization of one-shot segmentation results for AMOS (Ji et al., 2022) (row 1), BTCV (Landman et al., 2015) (row 2), FLARE’22 (Ma et al., [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 3 Pith papers

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

  1. SimCroP: Radiograph Representation Learning with Similarity-driven Cross-granularity Pre-training

    cs.CV 2025-09 conditional novelty 6.0 of 10

    SimCroP learns chest-CT representations by aligning each report sentence to its most similar visual patches and fusing whole-scan and word-patch features, reporting higher classification and segmentation scores than s...

  2. Pre-Trained LLM is a Semantic-Aware and Generalizable Segmentation Booster

    cs.CV 2025-06 conditional novelty 6.0 of 10

    A frozen pre-trained LLM layer inserted between a CNN encoder and decoder improves medical image segmentation across ultrasound, dermoscopy, polyp, and CT benchmarks with few added trainable parameters.

  3. U-RWKV: Lightweight medical image segmentation with direction-adaptive RWKV

    eess.IV 2025-07 conditional novelty 5.0 of 10

    U-RWKV is a lightweight U-shaped medical image segmenter that combines multi-directional RWKV scanning with stage-adaptive channel recalibration, reporting competitive Dice scores with about three million parameters.

Reference graph

Works this paper leans on

66 extracted references · 55 canonical work pages · cited by 3 Pith papers

  1. [1]

    , author Balestriero, R

    author Assran, M. , author Balestriero, R. , author Duval, Q. , author Bordes, F. , author Misra, I. , author Bojanowski, P. , author Vincent, P. , author Rabbat, M. , author Ballas, N. , year 2022 a. title The hidden uniform cluster prior in self-supervised learning . journal arXiv preprint arXiv:2210.07277

    arXiv 2022

  2. [2]

    , author Caron, M

    author Assran, M. , author Caron, M. , author Misra, I. , author Bojanowski, P. , author Bordes, F. , author Vincent, P. , author Joulin, A. , author Rabbat, M. , author Ballas, N. , year 2022 b. title Masked siamese networks for label-efficient learning , in: booktitle European Conference on Computer Vision , organization Springer . pp. pages 456--473

    work page 2022

  3. [3]

    , author Ghodasara, S

    author Baid, U. , author Ghodasara, S. , author Mohan, S. , author Bilello, M. , author Calabrese, E. , author Colak, E. , author Farahani, K. , author Kalpathy-Cramer, J. , author Kitamura, F.C. , author Pati, S. , et al., year 2021 . title The rsna-asnr-miccai brats 2021 benchmark on brain tumor segmentation and radiogenomic classification . journal arX...

    arXiv 2021

  4. [4]

    , author Dong, L

    author Bao, H. , author Dong, L. , author Piao, S. , author Wei, F. , year 2021 . title Beit: Bert pre-training of image transformers . journal arXiv preprint arXiv:2106.08254

    arXiv 2021

  5. [5]

    , author Christ, P

    author Bilic, P. , author Christ, P. , author Li, H.B. , author Vorontsov, E. , author Ben-Cohen, A. , author Kaissis, G. , author Szeskin, A. , author Jacobs, C. , author Mamani, G.E.H. , author Chartrand, G. , et al., year 2023 . title The liver tumor segmentation benchmark (lits) . journal Medical Image Analysis volume 84 , pages 102680

    work page 2023

  6. [6]

    , author Touvron, H

    author Caron, M. , author Touvron, H. , author Misra, I. , author J \'e gou, H. , author Mairal, J. , author Bojanowski, P. , author Joulin, A. , year 2021 . title Emerging properties in self-supervised vision transformers , in: booktitle Proceedings of the IEEE/CVF international conference on computer vision , pp. pages 9650--9660

    2021

  7. [7]

    , author Kornblith, S

    author Chen, T. , author Kornblith, S. , author Norouzi, M. , author Hinton, G. , year 2020 . title A simple framework for contrastive learning of visual representations , in: booktitle International conference on machine learning , organization PMLR . pp. pages 1597--1607

    work page 2020

  8. [8]

    , author Ding, M

    author Chen, X. , author Ding, M. , author Wang, X. , author Xin, Y. , author Mo, S. , author Wang, Y. , author Han, S. , author Luo, P. , author Zeng, G. , author Wang, J. , year 2024 a. title Context autoencoder for self-supervised representation learning . journal International Journal of Computer Vision volume 132 , pages 208--223

    work page 2024

Show all 66 references
  1. [9]

    , author Ding, M

    author Chen, X. , author Ding, M. , author Wang, X. , author Xin, Y. , author Mo, S. , author Wang, Y. , author Han, S. , author Luo, P. , author Zeng, G. , author Wang, J. , year 2024 b. title Context autoencoder for self-supervised representation learning . journal Internati...

    2024

  2. [10]

    , author Agarwal, D

    author Chen, Z. , author Agarwal, D. , author Aggarwal, K. , author Safta, W. , author Balan, M.M. , author Brown, K. , year 2023 a. title Masked image modeling advances 3d medical image analysis , in: booktitle Proceedings of the IEEE/CVF Winter Conference on Applications of ...

    2023

  3. [11]

    , author Agarwal, D

    author Chen, Z. , author Agarwal, D. , author Aggarwal, K. , author Safta, W. , author Balan, M.M. , author Brown, K. , year 2023 b. title Masked image modeling advances 3d medical image analysis , in: booktitle Proceedings of the IEEE/CVF Winter Conference on Applications of ...

    2023

  4. [12]

    , author Vendt, B

    author Clark, K. , author Vendt, B. , author Smith, K. , author Freymann, J. , author Kirby, J. , author Koppel, P. , author Moore, S. , author Phillips, S. , author Maffitt, D. , author Pringle, M. , et al., year 2013 . title The cancer imaging archive (tcia): maintaining and...

    2013

  5. [13]

    , author Bao, J

    author Dong, X. , author Bao, J. , author Zhang, T. , author Chen, D. , author Zhang, W. , author Yuan, L. , author Chen, D. , author Wen, F. , author Yu, N. , author Guo, B. , year 2023 . title Peco: Perceptual codebook for bert pre-training of vision transformers , in: bookt...

    2023

  6. [14]

    , year 2020

    author Dosovitskiy, A. , year 2020 . title An image is worth 16x16 words: Transformers for image recognition at scale . journal arXiv preprint arXiv:2010.11929

    2020 arXiv

  7. [15]

    , author Soboleva, V

    author Goncharov, M. , author Soboleva, V. , author Kurmukov, A. , author Pisov, M. , author Belyaev, M. , year 2023 . title vox2vec: A framework for self-supervised contrastive learning of voxel-level representations in medical images , in: booktitle International Conference ...

    2023

  8. [16]

    , author Clifton, D.A

    author Gowda, S.N. , author Clifton, D.A. , year 2024 . title Masks and manuscripts: Advancing medical pre-training with end-to-end masking and narrative structuring , in: booktitle International Conference on Medical Image Computing and Computer-Assisted Intervention , organi...

    2024

  9. [17]

    , author Taher, M.R.H

    author Haghighi, F. , author Taher, M.R.H. , author Zhou, Z. , author Gotway, M.B. , author Liang, J. , year 2021 . title Transferable visual words: Exploiting the semantics of anatomical patterns for self-supervised learning . journal IEEE transactions on medical imaging volu...

    2021

  10. [18]

    , author Tang, Y

    author Hatamizadeh, A. , author Tang, Y. , author Nath, V. , author Yang, D. , author Myronenko, A. , author Landman, B. , author Roth, H.R. , author Xu, D. , year 2022 . title Unetr: Transformers for 3d medical image segmentation , in: booktitle Proceedings of the IEEE/CVF wi...

    2022

  11. [19]

    , author Chen, X

    author He, K. , author Chen, X. , author Xie, S. , author Li, Y. , author Doll \'a r, P. , author Girshick, R. , year 2022 . title Masked autoencoders are scalable vision learners , in: booktitle Proceedings of the IEEE/CVF conference on computer vision and pattern recognition...

    2022

  12. [20]

    , author Fan, H

    author He, K. , author Fan, H. , author Wu, Y. , author Xie, S. , author Girshick, R. , year 2020 a. title Momentum contrast for unsupervised visual representation learning , in: booktitle Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pp. ...

    2020

  13. [21]

    , author Yang, G

    author He, Y. , author Yang, G. , author Ge, R. , author Chen, Y. , author Coatrieux, J.L. , author Wang, B. , author Li, S. , year 2023 . title Geometric visual similarity learning in 3d medical image self-supervised pre-training , in: booktitle Proceedings of the IEEE/CVF Co...

    2023

  14. [22]

    , author Yang, G

    author He, Y. , author Yang, G. , author Yang, J. , author Chen, Y. , author Kong, Y. , author Wu, J. , author Tang, L. , author Zhu, X. , author Dillenseger, J.L. , author Shao, P. , et al., year 2020 b. title Dense biased networks with deep priori anatomy and hard region ada...

    2020

  15. [23]

    , author Yang, G

    author He, Y. , author Yang, G. , author Yang, J. , author Ge, R. , author Kong, Y. , author Zhu, X. , author Zhang, S. , author Shao, P. , author Shu, H. , author Dillenseger, J.L. , et al., year 2021 . title Meta grayscale adaptive network for 3d integrated renal structures ...

    2021

  16. [24]

    , author Van Ginneken, B

    author Heimann, T. , author Van Ginneken, B. , author Styner, M.A. , author Arzhaeva, Y. , author Aurich, V. , author Bauer, C. , author Beck, A. , author Becker, C. , author Beichel, R. , author Bekes, G. , et al., year 2009 . title Comparison and evaluation of methods for li...

    2009

  17. [25]

    , author Li, C

    author Huang, W. , author Li, C. , author Zhou, H.Y. , author Yang, H. , author Liu, J. , author Liang, Y. , author Zheng, H. , author Zhang, S. , author Wang, S. , year 2024 . title Enhancing representation in radiography-reports foundation model: A granular alignment algorit...

    2024

  18. [26]

    , author Wald, T

    author Isensee, F. , author Wald, T. , author Ulrich, C. , author Baumgartner, M. , author Roy, S. , author Maier-Hein, K. , author Jaeger, P.F. , year 2024 . title nnu-net revisited: A call for rigorous validation in 3d medical image segmentation , in: booktitle International...

    2024

  19. [27]

    , author Bai, H

    author Ji, Y. , author Bai, H. , author Ge, C. , author Yang, J. , author Zhu, Y. , author Zhang, R. , author Li, Z. , author Zhanng, L. , author Ma, W. , author Wan, X. , et al., year 2022 . title Amos: A large-scale abdominal multi-organ benchmark for versatile medical image...

    2022

  20. [28]

    , author Gezer, N.S

    author Kavur, A.E. , author Gezer, N.S. , author Bar s , M. , author Aslan, S. , author Conze, P.H. , author Groza, V. , author Pham, D.D. , author Chatterjee, S. , author Ernst, P. , author \"O zkan, S. , et al., year 2021 . title Chaos challenge-combined (ct-mr) healthy abdo...

    2021

  21. [29]

    , author Xu, Z

    author Landman, B. , author Xu, Z. , author Igelsias, J. , author Styner, M. , author Langerak, T. , author Klein, A. , year 2015 . title Miccai multi-atlas labeling beyond the cranial vault--workshop and challenge , in: booktitle Proc. MICCAI Multi-Atlas Labeling Beyond Crani...

    2015

  22. [30]

    , author Birkbeck, N

    author Lay, N. , author Birkbeck, N. , author Zhang, J. , author Kevin Zhou, S. , year 2013 . title Rapid multi-organ segmentation using context integration and discriminative models , in: booktitle International Conference on Information Processing in Medical Imaging , organi...

    2013

  23. [31]

    , author Zhang, L

    author Li, S. , author Zhang, L. , author Wang, Z. , author Wu, D. , author Wu, L. , author Liu, Z. , author Xia, J. , author Tan, C. , author Liu, Y. , author Sun, B. , et al., year 2023 . title Masked modeling for self-supervised representation learning on vision and beyond ...

    2023 arXiv

  24. [32]

    , author Qu, C

    author Li, W. , author Qu, C. , author Chen, X. , author Bassi, P.R. , author Shi, Y. , author Lai, Y. , author Yu, Q. , author Xue, H. , author Chen, Y. , author Lin, X. , et al., year 2024 . title Abdomenatlas: A large-scale, detailed-annotated, & multi-center dataset for ef...

    2024

  25. [33]

    , author Liao, W

    author Luo, X. , author Liao, W. , author Xiao, J. , author Chen, J. , author Song, T. , author Zhang, X. , author Li, K. , author Metaxas, D.N. , author Wang, G. , author Zhang, S. , year 2022 . title Word: A large scale dataset, benchmark and clinical applicable study for ab...

    2022

  26. [34]

    , author Wang, Y

    author Ma, J. , author Wang, Y. , author An, X. , author Ge, C. , author Yu, Z. , author Chen, J. , author Zhu, Q. , author Dong, G. , author He, J. , author He, Z. , et al., year 2021 a. title Toward data-efficient learning: A benchmark for covid-19 ct lung and infection segm...

    2021

  27. [35]

    , author Zhang, Y

    author Ma, J. , author Zhang, Y. , author Gu, S. , author Ge, C. , author Ma, S. , author Young, A. , author Zhu, C. , author Meng, K. , author Yang, X. , author Huang, Z. , et al., year 2023 . title Unleashing the strengths of unlabeled data in pan-cancer abdominal organ quan...

    2023 arXiv

  28. [36]

    , author Zhang, Y

    author Ma, J. , author Zhang, Y. , author Gu, S. , author Zhu, C. , author Ge, C. , author Zhang, Y. , author An, X. , author Wang, C. , author Wang, Q. , author Liu, X. , et al., year 2021 b. title Abdomenct-1k: Is abdominal organ segmentation a solved problem? journal IEEE T...

    2021

  29. [37]

    , author Jung, C

    author Park, G.Y. , author Jung, C. , author Lee, S. , author Ye, J.C. , author Lee, S.W. , year 2024 . title Self-supervised debiasing using low rank regularization , in: booktitle Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pp. pages 1...

    2024

  30. [38]

    , author Krahenbuhl, P

    author Pathak, D. , author Krahenbuhl, P. , author Donahue, J. , author Darrell, T. , author Efros, A.A. , year 2016 . title Context encoders: Feature learning by inpainting , in: booktitle Proceedings of the IEEE conference on computer vision and pattern recognition , pp. pag...

    2016

  31. [39]

    , author Strojan, P

    author Podobnik, G. , author Strojan, P. , author Peterlin, P. , author Ibragimov, B. , author Vrtovec, T. , year 2023 . title Han-seg: The head and neck organ-at-risk ct and mr segmentation dataset . journal Medical physics volume 50 , pages 1917--1927

    2023

  32. [40]

    , author Yi, D

    author Rister, B. , author Yi, D. , author Shivakumar, K. , author Nobashi, T. , author Rubin, D.L. , year 2020 . title Ct-org, a new dataset for multiple organ segmentation in computed tomography . journal Scientific Data volume 7 , pages 381

    2020

  33. [41]

    , author Fischer, P

    author Ronneberger, O. , author Fischer, P. , author Brox, T. , year 2015 . title U-net: Convolutional networks for biomedical image segmentation , in: booktitle Medical image computing and computer-assisted intervention--MICCAI 2015: 18th international conference, Munich, Ger...

    2015

  34. [42]

    , author Farag, A

    author Roth, H. , author Farag, A. , author Turkbey, E.B. , author Lu, L. , author Liu, J. , author Summers, R.M. , year 2016 . title Data from pancreas-ct. journal The Cancer Imaging Archive

    2016

  35. [43]

    , author Vetterli, M

    author Roy, O. , author Vetterli, M. , year 2007 . title The effective rank: A measure of effective dimensionality , in: booktitle 2007 15th European signal processing conference , organization IEEE . pp. pages 606--610

    2007

  36. [44]

    , author Koehler, G

    author Roy, S. , author Koehler, G. , author Ulrich, C. , author Baumgartner, M. , author Petersen, J. , author Isensee, F. , author Jaeger, P.F. , author Maier-Hein, K.H. , year 2023 . title Mednext: transformer-driven scaling of convnets for medical image segmentation , in: ...

    2023

  37. [45]

    , author Traverso, A

    author Setio, A.A.A. , author Traverso, A. , author De Bel, T. , author Berens, M.S. , author Van Den Bogaard, C. , author Cerello, P. , author Chen, H. , author Dou, Q. , author Fantacci, M.E. , author Geurts, B. , et al., year 2017 . title Validation, comparison, and combina...

    2017

  38. [46]

    , author Qin, C

    author Shao, P. , author Qin, C. , author Yin, C. , author Meng, X. , author Ju, X. , author Li, J. , author Lv, Q. , author Zhang, W. , author Xu, Z. , year 2011 . title Laparoscopic partial nephrectomy with segmental renal artery clamping: technique and clinical outcomes . j...

    2011

  39. [47]

    , author Tang, L

    author Shao, P. , author Tang, L. , author Li, P. , author Xu, Y. , author Qin, C. , author Cao, Q. , author Ju, X. , author Meng, X. , author Lv, Q. , author Li, J. , et al., year 2012 . title Precise segmental renal artery clamping under the guidance of dual-source computed ...

    2012

  40. [48]

    , author Loetzsch, W

    author Taleb, A. , author Loetzsch, W. , author Danz, N. , author Severin, J. , author Gaertner, T. , author Bergner, B. , author Lippert, C. , year 2020 . title 3d self-supervised methods for medical imaging . journal Advances in neural information processing systems volume 3...

    2020

  41. [49]

    , author Xu, R

    author Tang, F. , author Xu, R. , author Yao, Q. , author Fu, X. , author Quan, Q. , author Zhu, H. , author Liu, Z. , author Zhou, S.K. , year 2024 . title Hyspark: Hybrid sparse masking for large scale medical image pre-training , in: booktitle International Conference on Me...

    2024

  42. [50]

    , author Yang, D

    author Tang, Y. , author Yang, D. , author Li, W. , author Roth, H.R. , author Landman, B. , author Xu, D. , author Nath, V. , author Hatamizadeh, A. , year 2022 . title Self-supervised pre-training of swin transformers for 3d medical image analysis , in: booktitle Proceedings...

    2022

  43. [51]

    , author Li, Y

    author Tao, X. , author Li, Y. , author Zhou, W. , author Ma, K. , author Zheng, Y. , year 2020 . title Revisiting rubik’s cube: Self-supervised learning with volume-wise transformation for 3d medical image segmentation , in: booktitle Medical Image Computing and Computer Assi...

    2020

  44. [52]

    , author Jiang, Y

    author Tian, K. , author Jiang, Y. , author Diao, Q. , author Lin, C. , author Wang, L. , author Yuan, Z. , year 2023 . title Designing bert for convolutional networks: Sparse and hierarchical masked modeling . journal arXiv preprint arXiv:2301.03580

    2023 arXiv

  45. [53]

    , author Zhuang, J

    author Wu, L. , author Zhuang, J. , author Chen, H. , year 2024 . title Voco: A simple-yet-effective volume contrastive learning framework for 3d medical image analysis , in: booktitle Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pp. page...

    2024

  46. [54]

    , author Zhang, J

    author Xie, Y. , author Zhang, J. , author Xia, Y. , author Wu, Q. , year 2022 a. title Unimiss: Universal medical self-supervised learning via breaking dimensionality barrier , in: booktitle European Conference on Computer Vision , organization Springer . pp. pages 558--575

    2022

  47. [55]

    , author Zhang, Z

    author Xie, Z. , author Zhang, Z. , author Cao, Y. , author Lin, Y. , author Bao, J. , author Yao, Z. , author Dai, Q. , author Hu, H. , year 2022 b. title Simmim: A simple framework for masked image modeling , in: booktitle Proceedings of the IEEE/CVF conference on computer v...

    2022

  48. [56]

    , author Li, X

    author Yang, H. , author Li, X. , author Tang, S. , author Zhu, F. , author Wang, Y. , author Chen, M. , author Bai, L. , author Zhao, R. , author Ouyang, W. , year 2022 . title Cycle-consistent masked autoencoder for unsupervised domain generalization , in: booktitle The Elev...

    2022

  49. [57]

    , author Zheng, H

    author Zhang, C. , author Zheng, H. , author Gu, Y. , year 2023 . title Dive into the details of self-supervised learning for medical image analysis . journal Medical Image Analysis volume 89 , pages 102879

    2023

  50. [58]

    , author Wang, Y

    author Zhang, Q. , author Wang, Y. , author Wang, Y. , year 2022 a. title How mask matters: Towards theoretical understandings of masked autoencoders . journal Advances in Neural Information Processing Systems volume 35 , pages 27127--27139

    2022

  51. [59]

    , author Wang, Y

    author Zhang, Q. , author Wang, Y. , author Wang, Y. , year 2022 b. title How mask matters: Towards theoretical understandings of masked autoencoders . journal Advances in Neural Information Processing Systems volume 35 , pages 27127--27139

    2022

  52. [60]

    , author Wang, Y

    author Zhang, Q. , author Wang, Y. , author Wang, Y. , year 2022 c. title How mask matters: Towards theoretical understandings of masked autoencoders . journal Advances in Neural Information Processing Systems volume 35 , pages 27127--27139

    2022

  53. [61]

    , author Tan, Z

    author Zhang, Y. , author Tan, Z. , author Yang, J. , author Huang, W. , author Yuan, Y. , . title Matrix information theory for self-supervised learning , in: booktitle Forty-first International Conference on Machine Learning

  54. [62]

    , author Lu, C

    author Zhou, H.Y. , author Lu, C. , author Chen, C. , author Yang, S. , author Yu, Y. , year 2023 a. title A unified visual information preservation framework for self-supervised pre-training in medical image analysis . journal IEEE Transactions on Pattern Analysis and Machine...

    2023

  55. [63]

    , author Liu, H

    author Zhou, L. , author Liu, H. , author Bae, J. , author He, J. , author Samaras, D. , author Prasanna, P. , year 2023 b. title Self pre-training with masked autoencoders for medical image classification and segmentation , in: booktitle 2023 IEEE 20th International Symposium...

    2023

  56. [64]

    , author Greenspan, H

    author Zhou, S.K. , author Greenspan, H. , author Shen, D. , year 2023 c. title Deep learning for medical image analysis . publisher Academic Press

    2023

  57. [65]

    , author Sodha, V

    author Zhou, Z. , author Sodha, V. , author Pang, J. , author Gotway, M.B. , author Liang, J. , year 2021 . title Models genesis . journal Medical image analysis volume 67 , pages 101840

    2021

  58. [66]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.