Motion-Guided Causal Disentanglement for Robust Multi-View Cine Cardiac MRI Diagnosis

Amit Patel; Chuankai Xu; Cristiane De Carvalho Singulane; Diego Perez de Arenaza; Fabio Fernandes; Jane Cao; Jeremy Slivnick; Jianxin Xie; Karolina Zareba; Mohammad Abuannadi

arxiv: 2606.04414 · v1 · pith:TFGJQJ6Bnew · submitted 2026-06-03 · 💻 cs.CV · cs.MM

Motion-Guided Causal Disentanglement for Robust Multi-View Cine Cardiac MRI Diagnosis

Chuankai Xu , Cristiane De Carvalho Singulane , Mohammad Abuannadi , Stephen Chandler , Jeremy Slivnick , Karolina Zareba , Jane Cao , Vidya Nadig

show 5 more authors

Fabio Fernandes Seth Uretsky Diego Perez de Arenaza Amit Patel Jianxin Xie

This is my paper

Pith reviewed 2026-06-28 06:45 UTC · model grok-4.3

classification 💻 cs.CV cs.MM

keywords cardiac MRIdisentanglementcontrastive learningmulti-view imagingmotion guidanceadversarial trainingdisease classification

0 comments

The pith

A motion-guided framework disentangles view-specific anatomy from disease features in multi-view cardiac MRI using dual contrastive objectives and adversarial constraints.

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

The paper aims to prevent models from mixing imaging-view differences with actual disease signals when analyzing cardiac MRI scans from multiple angles. Current transformer approaches create unified embeddings that let view-dependent structures dominate decisions, worsening performance when data is scarce or conditions are rare. The proposed approach adds motion cues from frame-to-frame differences to focus on the heart, then splits the latent space so one part captures only view information while the other isolates disease patterns through contrastive pulls and an adversarial push that blocks disease information from leaking into the view branch. If successful, this separation would let classifiers rely on pathology rather than scanner angle or background, improving reliability on both classification and segmentation across clinical and public datasets.

Core claim

The MoViD framework explicitly factorizes latent representations into view-specific and disease-discriminative components using dual-branch supervised contrastive objectives and a gradient-reversal adversarial constraint that minimizes disease leakage into the view embedding. An annotation-free temporal motion feature derived from inter-frame difference maps localizes the beating heart region and suppresses background artifacts, with focal reweighting added to the contrastive loss to handle class imbalance.

What carries the argument

Motion-Guided View-Disease Disentanglement framework on a ViT-MAE backbone that splits representations via dual-branch supervised contrastive losses plus gradient-reversal adversarial training, guided by inter-frame difference maps.

If this is right

The model outperforms standard transformer baselines on disease classification and cardiac segmentation across a private venous thrombosis dataset and the M&Ms and M&Ms2 benchmarks.
Performance remains competitive with large-scale pretrained foundation models despite using a smaller backbone.
The approach reduces shortcut learning and view-dependent boundaries especially in low-data regimes for underrepresented conditions.
Focal reweighting within the contrastive loss mitigates the effect of class imbalance during training.

Where Pith is reading between the lines

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

The same motion-plus-adversarial split could be tested on other multi-view modalities such as ultrasound or CT where anatomical pose varies independently of pathology.
If the view embedding truly contains no disease signal, swapping view labels across patients should leave disease prediction unchanged, providing a direct check on leakage.
Extending the motion guidance to 3D cine volumes might further stabilize the heart localization when through-plane motion is present.

Load-bearing premise

Inter-frame difference maps can localize the heart and remove background without adding new biases or needing extra labels.

What would settle it

An ablation that removes the motion feature or the adversarial constraint and measures whether disease-classification accuracy falls back to standard transformer levels on the same test sets.

Figures

Figures reproduced from arXiv: 2606.04414 by Amit Patel, Chuankai Xu, Cristiane De Carvalho Singulane, Diego Perez de Arenaza, Fabio Fernandes, Jane Cao, Jeremy Slivnick, Jianxin Xie, Karolina Zareba, Mohammad Abuannadi, Seth Uretsky, Stephen Chandler, Vidya Nadig.

**Figure 2.** Figure 2: Segmentation comparison on M&Ms (SAX), M&Ms2 (SAX), and M&Ms2 (LAX 4C). [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗

**Figure 3.** Figure 3: GradCAM attention maps at three temporal frames ( [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

read the original abstract

Multi-view cardiac magnetic resonance (CMR) imaging provides complementary anatomical information and is widely used for noninvasive disease assessment. Recent transformer-based models have demonstrated strong representation learning capabilities for CMR analysis; however, they typically learn unified latent embeddings that entangle view-specific anatomical variations with disease-related features. Such entanglement biases classifiers toward structural attributes rather than view-invariant pathological patterns. This issue is exacerbated in low-data regimes, particularly for underrepresented cardiac conditions, where limited samples increase the susceptibility to shortcut learning and view-dependent decision boundaries. To address this, we propose a Motion-Guided View--Disease Disentanglement framework MoViD built upon a ViT-MAE backbone. The model explicitly factorizes latent representations into view-specific and disease-discriminative components using dual-branch supervised contrastive objectives and a gradient-reversal adversarial constraint that minimizes disease leakage into the view embedding. Additionally, an annotation-free temporal motion feature, derived from inter-frame difference maps, is introduced to localize the beating heart region and suppress background artifacts. A focal reweighting mechanism is incorporated into the contrastive loss to mitigate class imbalance. We evaluate the framework on a private clinical venous thrombosis dataset and two public benchmarks (M&Ms, M&Ms2). Across disease classification and cardiac segmentation tasks, our approach consistently outperforms standard transformer baselines and demonstrates competitive performance against large-scale pretrained foundation models, validating the efficacy of structural disentanglement in medical image analysis.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper describes a motion-guided disentanglement setup for multi-view CMR using ViT-MAE, contrastive branches, and adversarial training, but the abstract contains no numbers or ablations so the claims cannot be checked.

read the letter

The one thing to know is that this work targets view-disease entanglement in multi-view cine CMR by feeding inter-frame difference maps into a ViT-MAE backbone, then using dual-branch supervised contrastive losses plus gradient-reversal adversarial training to push disease information out of the view embedding. A focal reweighting term handles class imbalance.

It does tackle a real clinical problem: standard transformers mix scanner or view artifacts with pathology signals, which hurts generalization on rare conditions. The annotation-free motion cue is a reasonable way to focus on the beating heart without extra labels, and the overall architecture is a direct, coherent extension of existing contrastive and adversarial disentanglement tricks to this domain.

The soft spots are the missing evidence. The abstract states consistent outperformance on a private venous thrombosis set and on M&Ms/M&Ms2 for both classification and segmentation, yet gives no accuracies, no baselines, no statistical tests, and no ablation results. That makes it impossible to judge whether the motion feature actually suppresses background noise or whether it leaks patient motion or arrhythmia artifacts into the supposedly clean view embedding. The stress-test concern about difference maps failing on irregular or low-SNR sequences is therefore still open.

This paper is mainly for groups already working on transformer models for cardiac MRI who need concrete ideas for handling multi-view bias. A reader looking for new theoretical machinery will not find it; a reader looking for a worked example of motion-guided disentanglement might pick up the loss design.

I would send it to peer review if the full manuscript supplies the missing quantitative comparisons and checks the motion feature on noisy cases; without that material it is too thin to evaluate.

Referee Report

2 major / 1 minor

Summary. The paper proposes MoViD, a Motion-Guided View-Disease Disentanglement framework built on a ViT-MAE backbone for multi-view cine cardiac MRI. It factorizes latent representations into view-specific and disease-discriminative components via dual-branch supervised contrastive objectives and a gradient-reversal adversarial constraint to minimize disease leakage into view embeddings. An annotation-free temporal motion feature from inter-frame difference maps localizes the beating heart and suppresses artifacts, with focal reweighting for class imbalance. The framework is evaluated on a private venous thrombosis dataset and public M&Ms/M&Ms2 benchmarks, claiming consistent outperformance over standard transformer baselines on disease classification and cardiac segmentation tasks, with competitive results against large pretrained models.

Significance. If the claimed performance gains hold under rigorous validation, the work could contribute to more robust representation learning in multi-view CMR by reducing view-dependent biases and shortcut learning, particularly in low-data regimes for rare conditions. The structured use of motion guidance alongside contrastive and adversarial objectives offers a concrete mechanism for disentanglement that may generalize to other multi-view medical imaging tasks.

major comments (2)

[Abstract] Abstract: The central claim of consistent outperformance on disease classification and cardiac segmentation is load-bearing for the manuscript's contribution, yet the abstract (and visible description) supplies no quantitative results, error bars, statistical tests, baseline details, or ablation studies. This prevents verification of whether the disentanglement components deliver the asserted gains over transformer baselines.
[Abstract] Abstract: The annotation-free temporal motion feature derived from inter-frame difference maps is presented as localizing the beating heart region and guiding the view embedding without new biases or supervision. This assumption is load-bearing for the disentanglement pipeline and the adversarial constraint. However, difference maps are known to be sensitive to global patient motion, frame-rate variation, arrhythmias, or low SNR, which could allow spurious activations or background leakage; no validation, robustness analysis, or ablation of this component is described.

minor comments (1)

[Abstract] Abstract: The title refers to 'Causal Disentanglement' but the description relies on supervised contrastive losses and gradient reversal without stating causal assumptions, identifiability conditions, or interventions.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on the abstract and the motion feature component. We address each major comment below and will revise the manuscript to strengthen the presentation of results and component validation.

read point-by-point responses

Referee: [Abstract] Abstract: The central claim of consistent outperformance on disease classification and cardiac segmentation is load-bearing for the manuscript's contribution, yet the abstract (and visible description) supplies no quantitative results, error bars, statistical tests, baseline details, or ablation studies. This prevents verification of whether the disentanglement components deliver the asserted gains over transformer baselines.

Authors: We agree that the abstract would benefit from concrete metrics to support the claims. In the revised manuscript, we will add key quantitative results (e.g., classification accuracy gains and segmentation Dice improvements over ViT-MAE baselines on M&Ms and the private dataset) while keeping the abstract concise. Detailed error bars, statistical tests, and full ablations remain in the main text and supplementary material, as is standard for abstract length limits. revision: yes
Referee: [Abstract] Abstract: The annotation-free temporal motion feature derived from inter-frame difference maps is presented as localizing the beating heart region and guiding the view embedding without new biases or supervision. This assumption is load-bearing for the disentanglement pipeline and the adversarial constraint. However, difference maps are known to be sensitive to global patient motion, frame-rate variation, arrhythmias, or low SNR, which could allow spurious activations or background leakage; no validation, robustness analysis, or ablation of this component is described.

Authors: The referee correctly notes that inter-frame difference maps can be affected by non-cardiac factors. Our current experiments implicitly rely on the downstream contrastive and adversarial losses to focus on cardiac motion, but we acknowledge the lack of explicit robustness checks. We will add a dedicated ablation subsection with qualitative examples and quantitative localization metrics (e.g., overlap with annotated heart regions) under simulated motion and SNR variations in the revision. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper introduces an architectural framework (MoViD) with dual-branch contrastive losses, gradient-reversal adversarial training, and an inter-frame difference map for motion guidance. These are presented as design choices evaluated empirically on external datasets (private venous thrombosis, M&Ms, M&Ms2), without any equations, fitted parameters, or predictions that reduce by construction to the inputs themselves. No self-citation chains, uniqueness theorems, or ansatzes imported from prior author work are invoked as load-bearing justifications. The central claims rest on comparative performance metrics rather than self-referential definitions, making the derivation self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides no explicit free parameters, axioms, or invented entities; the framework is described at the level of objectives and constraints without detailing fitted values or background assumptions.

pith-pipeline@v0.9.1-grok · 5837 in / 1233 out tokens · 26207 ms · 2026-06-28T06:45:26.457717+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

18 extracted references · 3 canonical work pages

[1]

IEEE Transactions on Medical Imaging40(12), 3543–3554 (2021)

Campello, V .M., Gkontra, P., Izquierdo, C., Martin-Isla, C., Soez, A., Full, P.M., et al.: Multi- centre, multi-vendor and multi-disease cardiac segmentation: the M&Ms challenge. IEEE Transactions on Medical Imaging40(12), 3543–3554 (2021)

2021
[2]

In: ICLR (2021)

Dosovitskiy, A., Beyer, L., Kolesnikov, A., et al.: An image is worth 16x16 words: Trans- formers for image recognition at scale. In: ICLR (2021)

2021
[3]

Diagnostics12(1), 69 (2022)

Duffy, G., Vulliard, L., Defined, M., Itié-Hafez, S., Camarasa, R., Boccalini, S., Si- Mohamed, S., Chevalier, P., Defined2, J.F., Defined3, L.: Deep learning supplants visual analysis by experienced operators for the diagnosis of cardiac amyloidosis by cine-CMR. Diagnostics12(1), 69 (2022)

2022
[4]

arXiv preprint arXiv:2506.00679 (2025)

Fu, Y ., Bai, W., Yi, W., Manisty, C., Bhuva, A.N., Treibel, T.A., Moon, J.C., Clarkson, M.J., Davies, R.H., Hu, Y .: A versatile foundation model for cine cardiac magnetic resonance image analysis tasks. arXiv preprint arXiv:2506.00679 (2025)

work page arXiv 2025
[5]

In: Journal of Machine Learning Research

Ganin, Y ., Ustinova, E., Ajakan, H., Germain, P., Larochelle, H., Laviolette, F., Marchand, M., Lempitsky, V .: Domain-adversarial training of neural networks. In: Journal of Machine Learning Research. vol. 17, pp. 1–35 (2016)

2016
[6]

European Heart Journal42(16), 1554–1568 (2021)

Garcia-Pavia, P., Rapezzi, C., Adler, Y ., Arad, M., Basso, C., Brucato, A., Burazor, I., Caforio, A.L.P., Damy, T., Eriksson, U., et al.: Diagnosis and treatment of cardiac amyloido- sis: a position statement of the ESC working group on myocardial and pericardial diseases. European Heart Journal42(16), 1554–1568 (2021)

2021
[7]

In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

He, K., Chen, X., Xie, S., Li, Y ., Dollár, P., Girshick, R.: Masked autoencoders are scal- able vision learners. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 16000–16009 (2022)

2022
[8]

In: Pro- ceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)

He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. In: Pro- ceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). pp. 770–778 (2016)

2016
[9]

arXiv preprint arXiv:2410.01665 (2024)

Jacob, A.J., Borgohain, I., Chitiboi, T., Sharma, P., Comaniciu, D., Rueckert, D.: Towards a vision foundation model for comprehensive assessment of cardiac MRI. arXiv preprint arXiv:2410.01665 (2024)

work page arXiv 2024
[10]

In: Advances in Neural Information Processing Systems (NeurIPS)

Khosla, P., Teterwak, P., Wang, C., Sarna, A., Tian, Y ., Isola, P., Maschinot, A., Liu, C., Kr- ishnan, D.: Supervised contrastive learning. In: Advances in Neural Information Processing Systems (NeurIPS). vol. 33, pp. 18661–18673 (2020)

2020
[11]

In: Proceedings of the IEEE International Conference on Computer Vision (ICCV)

Lin, T.Y ., Goyal, P., Girshick, R., He, K., Dollár, P.: Focal loss for dense object detection. In: Proceedings of the IEEE International Conference on Computer Vision (ICCV). pp. 2980– 2988 (2017)

2017
[12]

In: International Confer- ence on Learning Representations (ICLR) (2019) 10 C

Loshchilov, I., Hutter, F.: Decoupled weight decay regularization. In: International Confer- ence on Learning Representations (ICLR) (2019) 10 C. Xu et al

2019
[13]

IEEE Journal of Biomed- ical and Health Informatics27(7), 3302–3313 (2023)

Martin-Isla, C., Campello, V .M., Izquierdo, C., Kushibar, K., et al.: Deep learning segmenta- tion of the right ventricle in cardiac MRI: the M&Ms-2 challenge. IEEE Journal of Biomed- ical and Health Informatics27(7), 3302–3313 (2023)

2023
[14]

Journal of Cardiovascular Magnetic Resonance22(1), 84 (2020)

Martini, N., Aimo, A., Barison, A., Della Latta, D., Vergaro, G., Aquaro, G.D., Ripoli, A., Emdin, M., Passino, C.: Deep learning to diagnose cardiac amyloidosis from cardiovascular magnetic resonance. Journal of Cardiovascular Magnetic Resonance22(1), 84 (2020)

2020
[15]

In: Medical Image Com- puting and Computer Assisted Intervention – MICCAI 2024

Robinet, L., Berjaoui, A., Kheil, Z., Cohen-Jonathan Moyal, E.: DRIM: Learning disentan- gled representations from incomplete multimodal healthcare data. In: Medical Image Com- puting and Computer Assisted Intervention – MICCAI 2024. LNCS, vol. 15003, pp. 163–

2024
[16]

In: International Conference on Medical image computing and computer- assisted intervention

Ronneberger, O., Fischer, P., Brox, T.: U-net: Convolutional networks for biomedical im- age segmentation. In: International Conference on Medical image computing and computer- assisted intervention. pp. 234–241. Springer (2015)

2015
[17]

In: Proceedings of the IEEE international conference on computer vision

Tran, D., Bourdev, L., Fergus, R., Torresani, L., Paluri, M.: Learning spatiotemporal features with 3d convolutional networks. In: Proceedings of the IEEE international conference on computer vision. pp. 4489–4497 (2015)

2015
[18]

arXiv preprint arXiv:2402.05079 , year=

Wang, Z., Zheng, J.Q., Zhang, Y ., Cui, G., Li, L.: Mamba-unet: Unet-like pure visual mamba for medical image segmentation. arXiv preprint arXiv:2402.05079 (2024)

work page arXiv 2024

[1] [1]

IEEE Transactions on Medical Imaging40(12), 3543–3554 (2021)

Campello, V .M., Gkontra, P., Izquierdo, C., Martin-Isla, C., Soez, A., Full, P.M., et al.: Multi- centre, multi-vendor and multi-disease cardiac segmentation: the M&Ms challenge. IEEE Transactions on Medical Imaging40(12), 3543–3554 (2021)

2021

[2] [2]

In: ICLR (2021)

Dosovitskiy, A., Beyer, L., Kolesnikov, A., et al.: An image is worth 16x16 words: Trans- formers for image recognition at scale. In: ICLR (2021)

2021

[3] [3]

Diagnostics12(1), 69 (2022)

Duffy, G., Vulliard, L., Defined, M., Itié-Hafez, S., Camarasa, R., Boccalini, S., Si- Mohamed, S., Chevalier, P., Defined2, J.F., Defined3, L.: Deep learning supplants visual analysis by experienced operators for the diagnosis of cardiac amyloidosis by cine-CMR. Diagnostics12(1), 69 (2022)

2022

[4] [4]

arXiv preprint arXiv:2506.00679 (2025)

Fu, Y ., Bai, W., Yi, W., Manisty, C., Bhuva, A.N., Treibel, T.A., Moon, J.C., Clarkson, M.J., Davies, R.H., Hu, Y .: A versatile foundation model for cine cardiac magnetic resonance image analysis tasks. arXiv preprint arXiv:2506.00679 (2025)

work page arXiv 2025

[5] [5]

In: Journal of Machine Learning Research

Ganin, Y ., Ustinova, E., Ajakan, H., Germain, P., Larochelle, H., Laviolette, F., Marchand, M., Lempitsky, V .: Domain-adversarial training of neural networks. In: Journal of Machine Learning Research. vol. 17, pp. 1–35 (2016)

2016

[6] [6]

European Heart Journal42(16), 1554–1568 (2021)

Garcia-Pavia, P., Rapezzi, C., Adler, Y ., Arad, M., Basso, C., Brucato, A., Burazor, I., Caforio, A.L.P., Damy, T., Eriksson, U., et al.: Diagnosis and treatment of cardiac amyloido- sis: a position statement of the ESC working group on myocardial and pericardial diseases. European Heart Journal42(16), 1554–1568 (2021)

2021

[7] [7]

In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

He, K., Chen, X., Xie, S., Li, Y ., Dollár, P., Girshick, R.: Masked autoencoders are scal- able vision learners. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 16000–16009 (2022)

2022

[8] [8]

In: Pro- ceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)

He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. In: Pro- ceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). pp. 770–778 (2016)

2016

[9] [9]

arXiv preprint arXiv:2410.01665 (2024)

Jacob, A.J., Borgohain, I., Chitiboi, T., Sharma, P., Comaniciu, D., Rueckert, D.: Towards a vision foundation model for comprehensive assessment of cardiac MRI. arXiv preprint arXiv:2410.01665 (2024)

work page arXiv 2024

[10] [10]

In: Advances in Neural Information Processing Systems (NeurIPS)

Khosla, P., Teterwak, P., Wang, C., Sarna, A., Tian, Y ., Isola, P., Maschinot, A., Liu, C., Kr- ishnan, D.: Supervised contrastive learning. In: Advances in Neural Information Processing Systems (NeurIPS). vol. 33, pp. 18661–18673 (2020)

2020

[11] [11]

In: Proceedings of the IEEE International Conference on Computer Vision (ICCV)

Lin, T.Y ., Goyal, P., Girshick, R., He, K., Dollár, P.: Focal loss for dense object detection. In: Proceedings of the IEEE International Conference on Computer Vision (ICCV). pp. 2980– 2988 (2017)

2017

[12] [12]

In: International Confer- ence on Learning Representations (ICLR) (2019) 10 C

Loshchilov, I., Hutter, F.: Decoupled weight decay regularization. In: International Confer- ence on Learning Representations (ICLR) (2019) 10 C. Xu et al

2019

[13] [13]

IEEE Journal of Biomed- ical and Health Informatics27(7), 3302–3313 (2023)

Martin-Isla, C., Campello, V .M., Izquierdo, C., Kushibar, K., et al.: Deep learning segmenta- tion of the right ventricle in cardiac MRI: the M&Ms-2 challenge. IEEE Journal of Biomed- ical and Health Informatics27(7), 3302–3313 (2023)

2023

[14] [14]

Journal of Cardiovascular Magnetic Resonance22(1), 84 (2020)

Martini, N., Aimo, A., Barison, A., Della Latta, D., Vergaro, G., Aquaro, G.D., Ripoli, A., Emdin, M., Passino, C.: Deep learning to diagnose cardiac amyloidosis from cardiovascular magnetic resonance. Journal of Cardiovascular Magnetic Resonance22(1), 84 (2020)

2020

[15] [15]

In: Medical Image Com- puting and Computer Assisted Intervention – MICCAI 2024

Robinet, L., Berjaoui, A., Kheil, Z., Cohen-Jonathan Moyal, E.: DRIM: Learning disentan- gled representations from incomplete multimodal healthcare data. In: Medical Image Com- puting and Computer Assisted Intervention – MICCAI 2024. LNCS, vol. 15003, pp. 163–

2024

[16] [16]

In: International Conference on Medical image computing and computer- assisted intervention

Ronneberger, O., Fischer, P., Brox, T.: U-net: Convolutional networks for biomedical im- age segmentation. In: International Conference on Medical image computing and computer- assisted intervention. pp. 234–241. Springer (2015)

2015

[17] [17]

In: Proceedings of the IEEE international conference on computer vision

Tran, D., Bourdev, L., Fergus, R., Torresani, L., Paluri, M.: Learning spatiotemporal features with 3d convolutional networks. In: Proceedings of the IEEE international conference on computer vision. pp. 4489–4497 (2015)

2015

[18] [18]

arXiv preprint arXiv:2402.05079 , year=

Wang, Z., Zheng, J.Q., Zhang, Y ., Cui, G., Li, L.: Mamba-unet: Unet-like pure visual mamba for medical image segmentation. arXiv preprint arXiv:2402.05079 (2024)

work page arXiv 2024