Tetrahedron-Net for Medical Image Registration

Dantong Shi; Jinhai Xiang; Qianru Han; Shuai Guo; Xiang Bai; Xinwei He

REVIEW 4 major objections 4 minor 13 references

Tetrahedron-Net for Medical Image Registration

T0 review · 4 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read The paper claims that appending a second decoder, which fuses encoder and first-decoder features, to any U-Net-like registration backbone yields consistent accuracy gains on LPBA40, IXI, and OASIS.

desk verdict A simple, plausibly useful second-decoder plug-in for registration U-Nets, but the paper never isolates the decoder's contribution from a pretraining advantage and lacks variance reporting. read the letter →

arxiv 2505.04380 v1 pith:EMOU3RZZ submitted 2025-05-07 eess.IV cs.CVcs.IR

Jinhai Xiang , Shuai Guo , Qianru Han , Dantong Shi , Xinwei He , Xiang Bai This is my paper · ORCID

classification eess.IVcs.CVcs.IR

keywords medicalimageregistrationdeformableU-Nettwo-leveldecoderskipconnectionsdeformationfieldunsupervisedlearningbrainMRI

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 tries to establish that the standard single-encoder, single-decoder U-Net leaves registration accuracy on the table, and that the cheap fix is a second decoder rather than more elaborate skip connections. Tetrahedron-Net keeps one shared encoder and decodes twice: the first decoder behaves like a normal U-Net decoder, and the second decoder concatenates features from both the encoder and the first decoder at every scale. On three brain MRI benchmarks (LPBA40, IXI, OASIS), the paper reports higher Dice overlap and lower fractions of non-smooth deformation-field voxels than VoxelMorph, ViT-V-Net, and TransMorph. The paper's practical claim is that the two-level decoder is a drop-in upgrade that improves any U-Net-like registration backbone, which would matter because it converts an architecture redesign into a simple append step.

What carries the argument

The load-bearing piece is the two-level decoder, a second decoder branch appended to an off-the-shelf encoder-decoder backbone. In the default U-UNet instantiation, Decoder 2 mirrors Decoder 1's structure but its $k$-th block takes as input the concatenation of the corresponding encoder feature, the previous Decoder-1 feature, and its own previous feature: $x^k_{\mathrm{dec2}} = \mathrm{Up}(\mathrm{CR}([x^{L-k}_{\mathrm{enc}}, x^{k-1}_{\mathrm{dec1}}, x^{k-1}_{\mathrm{dec2}}]))$. This turns registration into a coarse-to-fine refinement: Decoder 1 supplies a prior, and Decoder 2 fuses it with multi-scale encoder details to predict the final deformation field. The paper tests the same machinery with UNet++, UNet3+, and DenseUNet as second-decoder structures, reporting the best scores with the dense-block version.

What would settle it

Retrain each comparison model with the same two-stage pretraining schedule used for Tetrahedron-Net (encoder and first decoder first, then the full network) and compare Dice on LPBA40; if the gap essentially disappears, the second decoder is not the source of the gain.

Watch

Extended reading notes

Core claim

The paper's central claim is that a U-Net-style registration network produces better deformation fields when its decoder is replaced by two cooperative decoders sharing one encoder: the first decoder reconstructs multi-scale features exactly as in U-Net, and the second decoder refines that reconstruction by concatenating, at each scale, the encoder feature with the preceding outputs of both decoders. This makes the final prediction a coarse-to-fine refinement of the first decoder's result. The paper reports that using a DenseUNet-style second decoder gives the best configuration, reaching 0.681 average Dice on LPBA40 compared with 0.658 for the VoxelMorph baseline, and that the same appended-decoder recipe produces consistent gains when applied to VoxelMorph, ViT-V-Net, TransMorph, and TransMorph-bspl on LPBA40, IXI, and OASIS.

Load-bearing premise

For the reported gains to show what the paper claims, the comparison models must differ from Tetrahedron-Net only by the added decoder, but the proposed models were pretrained in two stages while the comparison models were not, so pretraining could be carrying part of the improvement.

Editorial extensions

If this is right

Because the appended decoder couples only through skip connections and the loss is unchanged, any U-Net-like registration network can adopt the two-level decoder with a minimal code change.
The reported gains hold for both convolutional (VoxelMorph) and transformer-based (ViT-V-Net, TransMorph) backbones, so the benefit does not depend on a particular encoder.
The internal design of the second decoder matters: a DenseUNet-style second decoder outperforms plain UNet, UNet++, and UNet3+ second decoders on LPBA40.
Adding more than two decoders continues to improve Dice but roughly doubles parameters per decoder, so the paper chooses two levels as the default trade-off.
Pretraining the encoder and first decoder before adding the second decoder adds a small but consistent gain across all second-decoder variants.

Reading between the lines

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

Editorial inference: because the appended decoder is agnostic to the encoder and loss function, the same recipe should transfer to other dense prediction tasks, such as medical image segmentation, which the authors list as future work.
Editorial inference: the reported comparisons do not isolate the decoder from the two-stage pretraining protocol, so a replication that pretrains single-decoder baselines identically is needed before the full improvement is attributed to Decoder 2.
Editorial inference: the LPBA40 evaluation uses one nine-volume test set with no error bars, so repeated runs with different seeds are needed to show whether the 0.6 to 2.3 point Dice differences are stable.
Editorial inference: the reported faster convergence with Decoder 2 suggests the second branch acts partly as an optimization aid, which could be separated from capacity effects by measuring epoch-wise accuracy in a low-data setting.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

The idea is genuinely simple and worth taking seriously: append a second decoder that takes skip connections from both the encoder and the first decoder, and let it predict the deformation field in a coarse-to-fine way. Applied to VoxelMorph, ViT-V-Net, and TransMorph, the reported Dice gains are consistent in direction across LPBA40, IXI, and OASIS. That is a real result, and the paper deserves credit for running proper ablations: it checks the encoder skip connections, the number of decoders, and the pretraining step separately. The DenseUNet-style Dec2 variant consistently looks best, and the integration code appears straightforward to reproduce.

The soft spots are proportionate but not fatal. The load-bearing one is the pretraining confound. The proposed models are pretrained as Enc+Dec1 and then fine-tuned with Dec2, while the baselines are not described as receiving any warm-start. The paper's own Table 1 shows pretraining alone buys 0.2–0.3 Dice on LPBA40. Several reported gains over baselines are in the 0.4–2.3 range, so a same-protocol comparison (pretrain the baselines too, or train Tetrahedron from scratch) is needed before attributing the entire gap to the two-level decoder. The stress-test note gets this right. Second, no standard deviations or repeated runs are reported anywhere; on a 9-volume LPBA40 test set, a 0.2-point Dice difference could easily be noise. Third, there is an internal inconsistency for VoxelMorph on LPBA40: %|J|<0 is 0.384 in Table 4 but 0.288 in Table 5. Likely a typo, but it needs fixing. Finally, DenseUNet was selected as the default Dec2 based on LPBA40 test performance, then carried to IXI and OASIS; that is post-hoc model selection on the test set, mild but worth acknowledging.

None of this sinks the paper. The architecture is well motivated, the ablations are mostly clean, and the reported gains, while modest, are not implausible. What is missing is a controlled experiment plus error bars. The paper is worth a serious referee: the reviewer should ask for same-protocol pretraining comparisons, variance reporting, and correction of the Table 4/5 inconsistency. I would bring it to a reading group as an example of a simple architectural variant with a clear empirical claim, and I would cite it if I were writing a registration-backbone paper. Send it to peer review.

Referee Report

4 major / 4 minor

Summary. The paper proposes Tetrahedron-Net, an extension of U-Net-like architectures for unsupervised 3D medical image registration. The central idea is to append a second decoder that receives skip connections from both the encoder and the first decoder, enabling a coarse-to-fine prediction of the deformation field. The authors instantiate three versions of the second decoder (UNet++, UNet3+, DenseUNet) and also integrate the two-level decoder into VoxelMorph, ViT-V-Net, TransMorph, and TransMorph-bspl. Experiments are reported on LPBA40, IXI, and OASIS, showing consistent improvements in Dice score and reductions in the percentage of non-positive Jacobian determinants. Ablation studies examine the impact of pretraining, encoder skip connections, the number of decoder levels, and the structure of the second decoder.

Significance. If the reported gains hold under a controlled comparison, the contribution is valuable: the two-level decoder is a simple, architecture-agnostic modification that can be bolted onto any U-Net-like registration network, and the paper demonstrates it on four public backbones across three datasets. The ablations are a strength, as they attempt to isolate the contribution of the second decoder and its skip connections. However, the current evaluation does not establish that the gains come from the decoder itself rather than from the pretraining schedule, and the lack of any variance or significance reporting makes the small absolute improvements difficult to interpret. The work is therefore potentially significant but not yet convincingly supported.

major comments (4)

[§4.2.1, §4.2.4, Tables 1, 4, 5] The main comparison is confounded by pretraining. Section 4.2.4 states that the proposed models are 'loaded with pre-trained Enc and Dec1', and Table 1 shows that this pretraining alone improves Dice by 0.2–0.3% on LPBA40 (e.g., U-UNet from 0.665 to 0.667). The baselines in Tables 4 and 5 (VoxelMorph, ViT-V-Net, TransMorph) are not described as receiving any equivalent warm-start. Since the reported gains of Tetrahedron-Net over these baselines are 1.0–2.3% (e.g., VoxelMorph 0.658 vs. U-DenseUNet 0.681 in Table 4), part or all of the advantage could be due to the pretraining stage rather than the second decoder. The authors should rerun the baselines under the same two-stage pretraining protocol, or train all models from scratch, to isolate the architectural contribution.
[Tables 4 and 5] No measures of uncertainty are reported anywhere. The LPBA40 test set has only 9 volumes, and the IXI and OASIS splits are also relatively small. The reported differences (e.g., 0.671 vs. 0.658 for U-UNet vs. VoxelMorph in Table 4) could easily fall within run-to-run variability. The paper should report mean and standard deviation over at least three independent training runs with different seeds, and ideally include paired statistical tests (e.g., Wilcoxon signed-rank) for the Dice comparisons. Without this, the central claim of 'consistent performance gains' is not statistically supported.
[Tables 4 and 5] There is an internal inconsistency in the reported %|J|<0 for VoxelMorph on LPBA40: Table 4 lists 0.384, while Table 5 lists 0.288 for the same dataset and baseline. Since the paper uses the Jacobian metric to claim improved regularity of deformation fields, this discrepancy undermines the reliability of that claim. The authors should clarify which value is correct and ensure that all tables are consistent.
[§4.2.4, §4.3] The choice of DenseUNet as the second decoder for the main results in Table 5 is made based on LPBA40 test-set performance (Section 4.2.4 states that DenseUNet 'achieves the best results'), and the default two-level decoder configuration is similarly selected based on LPBA40 (Section 4.2.3). This is a form of model selection on the test set, which can inflate results on LPBA40 and may not generalize. The authors should either select the Dec2 variant using a validation set or report results for all Dec2 variants on IXI and OASIS to demonstrate that the choice is not overfitted to LPBA40.

minor comments (4)

[§3.3, §3.4] The description of the UNet++-based second decoder says 'each decoder layer is connected to all preceding decoder layers', but the general equation in Section 3.2.3 (Eq. 5) only concatenates the immediately previous Dec1 and Dec2 features. The relation between the specific decoder structures and the general formulation should be clarified, either by giving separate equations for each variant or by stating that Eq. (5) is the simplest case.
[Table 3] The 'Level of decoder 1' entry reports DSC of 0.657, while the U-UNet baseline in Table 1 without pretraining is 0.665 and in Table 2 without encoder features is 0.667. The relationship between these numbers is not explained; if the ablation in Table 3 uses a different configuration (e.g., without pretraining or with different skip connections), this should be stated explicitly.
[§4.1.2] Please state the number of random seeds used, the hardware/software versions, and the approximate training time. No code or trained models are provided, so these details are important for reproducibility. The current description ('one NVIDIA RTX 3090 GPU') is insufficient for an independent replication.
[§5 'Limitations and future work'] The stated limitation ('validated only on medical image registration') is framed as a limitation, but the paper explicitly claims generality to other dense prediction tasks in the introduction; the more relevant limitation is the absence of a controlled ablation isolating the decoder's benefit from the pretraining schedule, as noted above. Consider rephrasing the limitations section to address the actual threats to validity.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the paper's claims are empirical architecture comparisons against external benchmarks, and no prediction is forced by definition or by self-citation.

full rationale

This is an empirical architecture study, not a derivation-based paper. The central claim, that a second decoder concatenating encoder and first-decoder features improves registration accuracy, is supported by direct comparisons against external baselines (VoxelMorph, ViT-V-Net, TransMorph) and by ablation experiments. Equations (3)-(5) define the encoder and two decoder branches, but the claimed improvement is not a logical consequence of these definitions; it is an empirical observation measured by Dice score and Jacobian determinants. There are no load-bearing self-citations, no imported uniqueness theorems, and no ansatz smuggled in via citations from the same authors. The paper does not rename a known result under new coordinates, and it does not fit a parameter and then relabel that fit as a prediction. The only notable concerns are experimental rather than circular: the DenseUNet variant of Dec2 was selected based on LPBA40 test performance and then reported on the same dataset, and the proposed models use a pretrained Enc/Dec1 warm-start while the baseline models are not described as receiving the same pretraining. These are threats to internal validity or fairness of comparison, not circularity, because no reported quantity reduces to its own input by construction. Accordingly, the circularity score is 0.

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

The paper introduces no new physical or theoretical entities; the Tetrahedron-Net is an architectural arrangement of standard layers. The ledger instead captures the protocol choices (pretraining, Dec2 selection, lambda) that carry the empirical claim, plus the standard assumptions of unsupervised registration evaluation.

free parameters (3)

Pretraining of Enc+Dec1 = True
The proposed models are pretrained on the training set before adding Dec2 (Sections 4.2.1, 4.2.4). This is a protocol choice, not a fitted number, but it is a degree of freedom that the baselines do not appear to share, and it could explain part of the gains.
Second-level decoder structure = DenseUNet
Three Dec2 variants were tested on LPBA40 and DenseUNet was selected for the main comparisons (Section 4.2.4). This is a post-hoc model choice that affects the headline numbers.
Regularization weight lambda = 1
The smoothness weight in the loss (Equation 6) is set to 1 for all experiments (Section 4.1.2); a single hand-chosen value, not a fitted parameter.

assumptions (3)

domain assumption NCC similarity plus diffusion regularization is a valid registration objective
The method optimizes Equation 6 with no anatomical supervision, relying on the common assumption that maximizing image similarity yields clinically meaningful alignments.
domain assumption The public benchmarks with FreeSurfer preprocessing are comparable across methods
Baseline numbers for NiftyReg, SyN, VoxelMorph, etc. are taken as comparable to the authors' runs, but the exact preprocessing and test splits for baselines are not fully specified (Sections 4.1.1, 4.3).
domain assumption Dice overlap on 9 test volumes (LPBA40) is a reliable performance signal
LPBA40 uses 9 test volumes with no error bars or significance tests, so the ranking could be within noise (Section 4.1.1).

how reviews work

0 comments

Cite this review

Pith. "Pith review of Tetrahedron-Net for Medical Image Registration." pith.science (2026). https://pith.science/paper/EMOU3RZZ

@misc{pith2026250504380,
  author       = {Pith},
  title        = {Pith review of: Tetrahedron-Net for Medical Image Registration},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EMOU3RZZ}},
  note         = {Machine review of arXiv:2505.04380}
}

read the original abstract

Medical image registration plays a vital role in medical image processing. Extracting expressive representations for medical images is crucial for improving the registration quality. One common practice for this end is constructing a convolutional backbone to enable interactions with skip connections among feature extraction layers. The de facto structure, U-Net-like networks, has attempted to design skip connections such as nested or full-scale ones to connect one single encoder and one single decoder to improve its representation capacity. Despite being effective, it still does not fully explore interactions with a single encoder and decoder architectures. In this paper, we embrace this observation and introduce a simple yet effective alternative strategy to enhance the representations for registrations by appending one additional decoder. The new decoder is designed to interact with both the original encoder and decoder. In this way, it not only reuses feature presentation from corresponding layers in the encoder but also interacts with the original decoder to corporately give more accurate registration results. The new architecture is concise yet generalized, with only one encoder and two decoders forming a ``Tetrahedron'' structure, thereby dubbed Tetrahedron-Net. Three instantiations of Tetrahedron-Net are further constructed regarding the different structures of the appended decoder. Our extensive experiments prove that superior performance can be obtained on several representative benchmarks of medical image registration. Finally, such a ``Tetrahedron'' design can also be easily integrated into popular U-Net-like architectures including VoxelMorph, ViT-V-Net, and TransMorph, leading to consistent performance gains.

Figures

Figures reproduced from arXiv: 2505.04380 by the authors.

**Figure 1.** Visualization of unregistered image pairs and the overlapping images. Overlapping the two images with different colors suggests that there are significant mismatches in the image pairs. During training, a convolutional neural network is trained end-to-end to output the registration parameters for a source image with respect to the target one. Once the data-driven training phase is completed, the registration paramet… view at source ↗

**Figure 2.** The architecture of the proposed Tetrahedron-Net registration network. The registration network used in the figure is the U-UNet network. Firstly, the fixed image 𝑓 and the moving image 𝑚 are concated in the channel dimension as the input. After the encoder extracts the features and then two decoders generate the deformation field 𝜙. Then the spatial transformation network (STN) uses the generated deformation field … view at source ↗

**Figure 3.** U-UNet++ is adopted as the second-level decoder. Each node is composed of convolution and ReLU layers. coarse-grained features to be captured, resulting in clearer boundary delineation and consequently enhancing overall accuracy [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Figures from the paper (6 more)

**Figure 4.** Figure 4: shows the second-level decoder structure in the form of UNet3+. As shown, each layer within Decoder 2 (Dec2) integrates feature maps from all scales of Decoder 1 (Dec1) as well as larger-scale feature maps from its own structure. Note that feature maps at the same scal…

**Figure 5.** Figure 5: The second-level decoder is designed by UDenseUNet. 3.4. Skip Connections The encoder features are fused using the skip connection mechanism, and for the different Dec2s mentioned above, we used different skip connections. For U-UNet and UDenseUNet, we directly use c…

**Figure 7.** Figure 7: shows the visualization of the results of the different registration methods on LPBA40. On the left side, it shows the input image pairs, fixed image (𝑓), and moving images (𝑚). The right side exhibits, from top to bottom, the deformed images, deformation fields images…

**Figure 6.** Figure 6: Training Loss and Validation Dice scores for LPBA40 during training [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

**Figure 8.** Figure 8: Visualization of registration results on datasets LPBA40, IXI. These are the images from baseline VoxelMorph, ViT-VNet, TransMorph (row 2), and their respective model trained with Dec2 (row 1) [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]

**Figure 9.** Figure 9: Validation Dice scores on LPBA40 and IXI during training with Dec2 on different registration models. LPBA40, IXI, and OAISIS, showing that the proposed Tetrahedron-Net can outperform the state-of-the-art methods by a large margin. Limitations and future work. our work…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

13 extracted references · 5 canonical work pages

[3]

U-resnet: Ultimate coupling of registration and seg- mentation with deep nets, in: Medical Image Computing and Computer Assisted Intervention–MICCAI 2019: 22nd International Conference, Shenzhen, China, October 13–17, 2019, Proceedings, Part III 22, pp. 310–319. Fischl, B.,

work page 2019
[5]

Intelligent Medicine 3, 59–78

Transformers in medical image analysis. Intelligent Medicine 3, 59–78. He,K.,Zhang,X.,Ren,S.,Sun,J.,2016. Deepresiduallearningforimage recognition,in:ProceedingsoftheIEEEconferenceoncomputervision and pattern recognition, pp. 770–778. Hu, M., Zhang, J., Matkovic, L., Liu, T., Yang, X.,

work page 2016
[8]

Medical ImageAnalysis85,102762

Transformingmedicalimagingwithtransformers?acomparativereview of key properties, current progresses, and future perspectives. Medical ImageAnalysis85,102762. doi: https://doi.org/10.1016/j.media.2023. 102762. Liu, Z., Lin, Y., Cao, Y., Hu, H., Wei, Y., Zhang, Z., Lin, S., Guo, B.,

doi:10.1016/j.media.2023 2023
[9]

arXivpreprint arXiv:1804.03999

Attentionu-net:Learningwheretolookforthepancreas. arXivpreprint arXiv:1804.03999 . Razzak, M.I., Naz, S., Zaib, A.,

arXiv
[10]

Classification in BioApps: Automation of decision making , 323–350

Deep learning for medical image processing: Overview, challenges and the future. Classification in BioApps: Automation of decision making , 323–350. Ronneberger,O.,Fischer,P.,Brox,T.,2015. U-net:Convolutionalnetworks for biomedical image segmentation, in: International Conference on Medical Image Computing and Computer-Assisted Intervention. Sheikhjafari,...

work page 2015
[12]

arXiv preprint arXiv:1809.03443

Inverse-consistent deep networks for unsupervised de- formable image registration. arXiv preprint arXiv:1809.03443 . 10 Zhang, J., Wang, J., Wang, X., Feng, D.,

arXiv
[97]

1088/0031-9155/59/1/97

URL: https://dx.doi.org/10.1088/0031-9155/59/1/97, doi:10. 1088/0031-9155/59/1/97. Zhao, A., Balakrishnan, G., Durand, F., Guttag, J.V., Dalca, A.V., 2019a. Dataaugmentationusinglearnedtransformsforone-shotmedicalimage segmentation. ArXiv abs/1902.09383. Zhao, S., Dong, Y., Chang, E.I., Xu, Y., et al., 2019b. Recursive cascaded networks for unsupervised m...

arXiv
[2014]

arXiv preprint arXiv:1412.6980

Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 . Kuang, D., Schmah, T.,

arXiv

Show all 13 references

[2015]

Going deeper with convolu- tions, in: Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 1–9. Wang, H., Cao, P., Wang, J., Zaiane, O.R., 2022a. Uctransnet: rethinking the skip connections in u-net from a channel-wise perspective with transformer...

2021
[2018]

ArXiv abs/1811.09243

Faim - a convnet method for unsupervised 3d medical image registration. ArXiv abs/1811.09243. Li, J., Chen, J., Tang, Y., Wang, C., Landman, B.A., Zhou, S.K.,

arXiv
[2019]

IEEE Transactions on medical imaging 38, 1788–1800

Voxelmorph: a learning framework for deformable medical image reg- istration. IEEE Transactions on medical imaging 38, 1788–1800. Chen,J.,Frey,E.C.,He,Y.,Segars,W.P.,Li,Y.,Du,Y.,2022. Transmorph: Transformer for unsupervised medical image registration. Medical image analysis 8...

2022 arXiv
[2020]

arXiv preprint arXiv:2010.11929

An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929 . Estienne, T., Vakalopoulou, M., Christodoulidis, S., Battistela, E., Ler- ousseau,M.,Carre,A.,Klausner,G.,Sun,R.,Robert,C.,Mougiakakou, S., et al.,

2010 arXiv
[2023]

arXiv preprint arXiv:2309.00727

Deep learning in medical image registration: introduction and survey. arXiv preprint arXiv:2309.00727 . He,K.,Gan,C.,Li,Z.,Rekik,I.,Yin,Z.,Ji,W.,Gao,Y.,Wang,Q.,Zhang, J., Shen, D.,

arXiv

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.

[1] [3]

U-resnet: Ultimate coupling of registration and seg- mentation with deep nets, in: Medical Image Computing and Computer Assisted Intervention–MICCAI 2019: 22nd International Conference, Shenzhen, China, October 13–17, 2019, Proceedings, Part III 22, pp. 310–319. Fischl, B.,

work page 2019

[2] [5]

Intelligent Medicine 3, 59–78

Transformers in medical image analysis. Intelligent Medicine 3, 59–78. He,K.,Zhang,X.,Ren,S.,Sun,J.,2016. Deepresiduallearningforimage recognition,in:ProceedingsoftheIEEEconferenceoncomputervision and pattern recognition, pp. 770–778. Hu, M., Zhang, J., Matkovic, L., Liu, T., Yang, X.,

work page 2016

[3] [8]

Medical ImageAnalysis85,102762

Transformingmedicalimagingwithtransformers?acomparativereview of key properties, current progresses, and future perspectives. Medical ImageAnalysis85,102762. doi: https://doi.org/10.1016/j.media.2023. 102762. Liu, Z., Lin, Y., Cao, Y., Hu, H., Wei, Y., Zhang, Z., Lin, S., Guo, B.,

doi:10.1016/j.media.2023 2023

[4] [9]

arXivpreprint arXiv:1804.03999

Attentionu-net:Learningwheretolookforthepancreas. arXivpreprint arXiv:1804.03999 . Razzak, M.I., Naz, S., Zaib, A.,

arXiv

[5] [10]

Classification in BioApps: Automation of decision making , 323–350

Deep learning for medical image processing: Overview, challenges and the future. Classification in BioApps: Automation of decision making , 323–350. Ronneberger,O.,Fischer,P.,Brox,T.,2015. U-net:Convolutionalnetworks for biomedical image segmentation, in: International Conference on Medical Image Computing and Computer-Assisted Intervention. Sheikhjafari,...

work page 2015

[6] [12]

arXiv preprint arXiv:1809.03443

Inverse-consistent deep networks for unsupervised de- formable image registration. arXiv preprint arXiv:1809.03443 . 10 Zhang, J., Wang, J., Wang, X., Feng, D.,

arXiv

[7] [97]

1088/0031-9155/59/1/97

URL: https://dx.doi.org/10.1088/0031-9155/59/1/97, doi:10. 1088/0031-9155/59/1/97. Zhao, A., Balakrishnan, G., Durand, F., Guttag, J.V., Dalca, A.V., 2019a. Dataaugmentationusinglearnedtransformsforone-shotmedicalimage segmentation. ArXiv abs/1902.09383. Zhao, S., Dong, Y., Chang, E.I., Xu, Y., et al., 2019b. Recursive cascaded networks for unsupervised m...

arXiv

[8] [2014]

arXiv preprint arXiv:1412.6980

Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 . Kuang, D., Schmah, T.,

arXiv