C2W-Tune: Cavity-to -Wall Transfer Learning for Thin Atrial Wall Segmentation in 3D LGE-MRI

Rebecca Thornhill; Sreeraman Rajan; Yusri Al-Sanaani

arxiv: 2603.24992 · v3 · submitted 2026-03-26 · 💻 cs.CV

C2W-Tune: Cavity-to -Wall Transfer Learning for Thin Atrial Wall Segmentation in 3D LGE-MRI

Yusri Al-Sanaani , Rebecca Thornhill , Sreeraman Rajan This is my paper

Pith reviewed 2026-05-15 00:44 UTC · model grok-4.3

classification 💻 cs.CV

keywords left atrial wall segmentation3D LGE-MRItransfer learningcavity segmentationthin wallU-Netfibrosis quantificationmedical image segmentation

0 comments

The pith

Transferring weights from a cavity segmentation model raises thin left atrial wall Dice from 0.623 to 0.814 in 3D LGE-MRI.

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

The paper introduces C2W-Tune, a two-stage method that first trains a 3D U-Net to segment the left atrial cavity and then transfers those weights to segment the much thinner atrial wall. The transfer uses progressive layer unfreezing to keep useful cavity features while refining for wall detail. This addresses the low contrast and thin geometry that make direct wall segmentation unreliable in late gadolinium-enhanced MRI. Readers care because wall thickness maps and fibrosis measurements guide treatment decisions in atrial fibrillation. Experiments on the 2018 LA Segmentation Challenge dataset show clear gains over an identical network trained from scratch on wall labels alone.

Core claim

C2W-Tune pre-trains a 3D U-Net with ResNeXt encoder and instance normalization on LA cavity segmentation to learn robust atrial representations, then transfers the weights to the wall task and adapts them with a progressive layer-unfreezing schedule. On the 2018 LA Segmentation Challenge dataset this produces a wall Dice of 0.814 versus 0.623 for the matched baseline, surface Dice at 1 mm of 0.731 versus 0.553, HD95 of 2.55 mm versus 2.95 mm, and ASSD of 0.63 mm versus 0.71 mm. The same transferred model reaches Dice 0.78 when trained on only 70 volumes.

What carries the argument

C2W-Tune, the cavity-to-wall transfer framework that pre-trains on cavity segmentation then applies progressive layer-unfreezing to adapt the same 3D U-Net for wall segmentation.

If this is right

Wall thickness mapping and fibrosis quantification become more reliable with the higher Dice and lower surface distances.
The approach stays competitive even when only 70 labeled wall volumes are available.
Progressive unfreezing preserves cavity features while allowing wall-specific refinement without major hyperparameter changes.
The method reduces dependence on large wall-specific annotation sets by reusing cavity models.

Where Pith is reading between the lines

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

The same cavity-to-wall transfer idea could be tested on other thin cardiac or vascular structures where cavity data are easier to obtain than wall labels.
An end-to-end multi-task model that jointly predicts cavity and wall might remove the need for staged fine-tuning.
If the cavity prior creates unwanted shape bias, adding a small adversarial term during fine-tuning could be checked as a simple extension.

Load-bearing premise

The cavity segmentation features remain useful anatomical priors for the wall task and do not introduce shape biases that hurt thin-wall accuracy during fine-tuning.

What would settle it

Training the identical 3D U-Net architecture from scratch on the full wall-labeled dataset and reaching wall Dice of 0.814 or higher would show that cavity pre-training adds no benefit.

read the original abstract

Accurate segmentation of the left atrial (LA) wall in 3D late gadolinium-enhanced MRI (LGE-MRI) is essential for wall thickness mapping and fibrosis quantification, yet it remains challenging due to the wall's thin geometry, complex anatomy, and low contrast. We propose C2W-Tune, a two-stage cavity-to-wall transfer framework that leverages a high-accuracy LA cavity model as an anatomical prior to improve thin-wall delineation. Using a 3D U-Net with a ResNeXt encoder and instance normalization, Stage 1 pre-trains the network to segment the LA cavity, learning robust atrial representations. Stage 2 transfers these weights and adapts the network to LA wall segmentation using a progressive layer-unfreezing schedule to preserve cavity features while enabling wall-specific refinement. On the 2018 LA Segmentation Challenge dataset, C2W-Tune outperformed an architecture-matched baseline trained from scratch. The wall Dice score increased from 0.623 to 0.814, surface Dice at 1 mm increased from 0.553 to 0.731, 95th-percentile Hausdorff distance (HD95) decreased from 2.95 mm to 2.55 mm, and average symmetric surface distance (ASSD) decreased from 0.71 mm to 0.63 mm. Under reduced supervision using 70 training volumes sampled from the same training set, C2W-Tune achieved a Dice of 0.78, remaining competitive with recent multi-class bi-atrial benchmarks, typically 0.6-0.7. These results show that anatomically grounded task transfer with controlled fine-tuning improves accuracy for thin LA wall segmentation in 3D LGE-MRI.

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

Cavity pretraining plus progressive unfreezing lifts LA wall Dice from 0.623 to 0.814 on the 2018 challenge set, but the design leaves open whether the gains come from the cavity prior or the fine-tuning schedule itself.

read the letter

The one thing to know is that this paper reports a solid lift in thin atrial wall segmentation performance by pre-training a 3D U-Net on left atrial cavity segmentation and then fine-tuning it for the wall task using a progressive layer-unfreezing schedule. On the 2018 LA Segmentation Challenge dataset, they move wall Dice from 0.623 to 0.814, surface Dice at 1mm from 0.553 to 0.731, and HD95 from 2.95mm to 2.55mm compared to an architecture-matched model trained from scratch. They also show it holds up reasonably when training data is cut to 70 volumes. The work does a few things right. The task is genuinely hard because the wall is thin and low-contrast in LGE-MRI, and using the cavity as a prior is a reasonable way to inject anatomical knowledge. The progressive unfreezing is a controlled way to adapt without destroying the pre-trained features. Reporting multiple surface-based metrics alongside Dice is appropriate for this geometry. The reduced-supervision experiment adds some practical value. The soft spot is the missing control experiment. The headline comparison pits the full C2W-Tune pipeline against a scratch baseline, but there is no run that applies the same progressive unfreezing and Stage-2 hyperparameters to a randomly initialized network. If that version already captures most of the gain, then the cavity pre-training is contributing less than it appears. The abstract is also thin on exact training details, split information, and any statistical testing, which makes it harder to assess how reliable the numbers are. No results on an independent test set beyond the challenge data. This is aimed at the cardiac MRI segmentation community, especially those focused on atrial fibrillation and fibrosis quantification. A reader working on similar transfer learning setups or thin-structure segmentation would get something out of the controlled adaptation schedule. I think it deserves peer review. The empirical results are concrete and the method is straightforward, so referees can check the claims and suggest the needed ablations.

Referee Report

2 major / 1 minor

Summary. The manuscript proposes C2W-Tune, a two-stage transfer learning framework for thin left atrial wall segmentation in 3D LGE-MRI. Stage 1 pre-trains a 3D U-Net with ResNeXt encoder and instance normalization on LA cavity segmentation; Stage 2 transfers the weights and fine-tunes for wall segmentation via progressive layer-unfreezing. On the 2018 LA Segmentation Challenge dataset, C2W-Tune improves wall Dice from 0.623 to 0.814, surface Dice at 1 mm from 0.553 to 0.731, HD95 from 2.95 mm to 2.55 mm, and ASSD from 0.71 mm to 0.63 mm over an architecture-matched scratch baseline, with competitive performance under reduced supervision (70 volumes, Dice 0.78).

Significance. If the gains hold after controlling for the fine-tuning protocol, the work demonstrates a practical way to inject anatomical priors from an easier cavity task into thin-wall segmentation, which is clinically relevant for wall-thickness mapping and fibrosis quantification. The evaluation on a public benchmark with multiple complementary surface metrics (Dice, surface Dice, HD95, ASSD) provides a solid empirical foundation.

major comments (2)

[Experimental results] Experimental results (comparison to baseline): the headline improvements (wall Dice 0.623→0.814) compare C2W-Tune (cavity pretrain + progressive unfreezing) against a scratch baseline, but no ablation applies the identical Stage-2 unfreezing schedule and hyperparameters to a randomly initialized network. Without this control, the contribution of transferable cavity priors cannot be isolated from the fine-tuning procedure itself.
[Methods] Methods section: training schedules, exact data splits, and statistical testing (e.g., significance of the reported metric differences) are not fully detailed, which prevents independent verification of the quantitative claims on the public dataset.

minor comments (1)

The abstract and results could explicitly state the number of volumes and any cross-validation strategy used in the reduced-supervision experiment for reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which help clarify the contributions of our work. We address each major point below and will revise the manuscript to incorporate the suggested improvements.

read point-by-point responses

Referee: [Experimental results] Experimental results (comparison to baseline): the headline improvements (wall Dice 0.623→0.814) compare C2W-Tune (cavity pretrain + progressive unfreezing) against a scratch baseline, but no ablation applies the identical Stage-2 unfreezing schedule and hyperparameters to a randomly initialized network. Without this control, the contribution of transferable cavity priors cannot be isolated from the fine-tuning procedure itself.

Authors: We agree that an ablation applying the identical Stage-2 progressive unfreezing schedule to a randomly initialized network is required to isolate the benefit of the cavity priors. In the revised manuscript we will add this control experiment and report the full set of metrics (Dice, surface Dice, HD95, ASSD) for direct comparison. revision: yes
Referee: [Methods] Methods section: training schedules, exact data splits, and statistical testing (e.g., significance of the reported metric differences) are not fully detailed, which prevents independent verification of the quantitative claims on the public dataset.

Authors: We will expand the Methods section to include the complete training schedules (learning rates, number of epochs per stage, batch sizes, and the precise progressive unfreezing protocol), the exact data splits used from the 2018 LA Segmentation Challenge, and statistical significance testing (paired t-tests with p-values) for all reported metric differences. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper presents an empirical transfer-learning pipeline (cavity pre-training followed by progressive unfreezing for wall segmentation) whose performance claims rest on direct numerical comparisons against an architecture-matched from-scratch baseline on the external 2018 LA Segmentation Challenge dataset. No equations, fitted parameters, or self-referential definitions appear that would reduce the reported Dice or surface-distance gains to quantities defined by the method itself. The central results are therefore independent of the method's own inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The method relies on standard deep-learning assumptions about U-Net feature reuse and transferability of cavity features to wall boundaries; no new entities or fitted constants beyond ordinary training hyperparameters are introduced.

axioms (1)

domain assumption Features learned for cavity segmentation remain useful priors for wall segmentation after controlled fine-tuning
Invoked in the description of Stage 2 transfer and progressive unfreezing

pith-pipeline@v0.9.0 · 5636 in / 1258 out tokens · 25143 ms · 2026-05-15T00:44:46.005523+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Foundation/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

two-stage cavity-to-wall transfer framework... progressive layer-unfreezing schedule
IndisputableMonolith/Foundation/AbsoluteFloorClosure.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

3D U-Net with ResNeXt encoder... DiceFocal loss

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

20 extracted references · 20 canonical work pages

[1]

Li, L., Zimmer, V.A., Schnabel, J.A., Zhuang, X.: Medical image analysis on left atrial LGE MRI for atrial fibrillation studies: a review. Med. Image Anal. 77, 102360 (2022). https://doi.org/10.1016/j.media.2022.102360

work page doi:10.1016/j.media.2022.102360 2022
[2]

A global benchmark of algorithms for segmenting the left atrium from late gadolinium-enhanced cardiac magnetic resonance imaging,

Xiong, Z. et al.: A global benchmark of algorithms for segmenting the left atrium from late gadolinium-enhanced cardiac magnetic resonance imaging. Med. Image Anal. 67, 101832 (2021). https://doi.org/10.1016/j.media.2020.101832

work page doi:10.1016/j.media.2020.101832 2021
[3]

Maceira, A.M., Cosín-Sales, J., Roughton, M., Prasad, S.K., Pennell, D.J.: Reference left atrial dimensions and volumes by steady state free precession cardiovascular magnetic resonance. J. Cardiovasc. Magn. Reson. 12, 65 (2010). https://doi.org/10.1186/1532- 429X-12-65

work page doi:10.1186/1532- 2010
[4]

et al.: A robust computational framework for estimating 3D bi-atrial chamber wall thickness

Wang, Y. et al.: A robust computational framework for estimating 3D bi-atrial chamber wall thickness. Comput. Biol. Med. 114, 103444 (2019). https://doi.org/10.1016/j.compbiomed.2019.103444

work page doi:10.1016/j.compbiomed.2019.103444 2019
[5]

Petitjean, C., Dacher, J.-N.: A review of segmentation methods in short axis cardiac MR images. Med. Image Anal. 15(2), 169– 184 (2011). https://doi.org/10.1016/j.media.2010.12.004

work page doi:10.1016/j.media.2010.12.004 2011
[6]

et al.: Left atrial segmentation in a few seconds using fully convolutional network and transfer learning

Puybareau, É. et al.: Left atrial segmentation in a few seconds using fully convolutional network and transfer learning . In: Pop, M. et al. (eds.) Statistical Atlases and Computational Models of the Heart. Atrial Segmentation and LV Quantification Challenges. LNCS, vol. 11395, pp. 339–347. Springer International, Cham (2019). https://doi.org/10.1007/978-...

work page doi:10.1007/978-3-030-12029-0_37 2019
[7]

Li, L., Zimmer, V.A., Schnabel, J.A., Zhuang, X.: AtrialJSQnet: a new framework for joint segmentation and quantification of left atrium and scars incorporating spatial and shape information. Med. Image Anal. 76, (2022). https://doi.org/10.1016/j.media.2021.102303

work page doi:10.1016/j.media.2021.102303 2022
[8]

An Ensemble of 3D Residual Encoder UNet Models for Solving Multi - class Bi-atrial Segmentation Challenge,

Zolotarev, A., Johnson, K., Khan, A., Slabaugh, G., Roney, C.: An ensemble of 3D residual encoder UNet models for solving multi-class bi-atrial segmentation challenge. In: Camara, O., et al. (eds.) Statistical Atlases and Computational Models of the Heart. Workshop, CMRxRecon and MBAS Challenge Papers. LNCS, vol. 15448, pp. 209 –213. Springer, Cham (2025)...

work page doi:10.1007/978-3-031-87756-8_21 2025
[9]

Evaluating Convolution, Attention, and Mamba Based U -Net Models for Multi- class Bi -Atrial Segmentation from LGE-MRI,

Thesing, C., Bueno-Orovio, A., Banerjee, A.: Evaluating convolution, attention, and Mamba based U-Net models for multi-class bi-atrial segmentation from LGE-MRI. In: Camara, O., et al. (eds.) Statistical Atlases and Computational Models of the Heart. Workshop, CMRxRecon and MBAS Challenge Papers. LNCS, vol. 15448, pp. 214 –225. Springer, Cham (2025). http...

work page doi:10.1007/978-3-031-87756-8_22 2025
[10]

Multi-loss 3D Segmentation for Enhanced Bi- atrial Segmentation,

Almar-Munoz, E. et al.: Multi -loss 3D segmentation for enhanced bi-atrial segmentation. In: Camara, O., et al. (eds.) Statistical Atlases and Computational Models of the Heart. Workshop, CMRxRecon and MBAS Challenge Papers. LNCS, vol. 15448, pp. 236–244. Springer, Cham (2025). https://doi.org/10.1007/978-3-031-87756-8_24

work page doi:10.1007/978-3-031-87756-8_24 2025
[11]

In: Proc

Zhang, Y., Cheng, Y., Qi, Y.: Exploring CNN and transformer architectures for multi -class bi -atrial segmentation from late gadolinium-enhanced MRI. In: Camara, O., et al. (eds.) Statistical Atlases and Computational Models of the Heart. Workshop, CMRxRecon and MBAS Challenge Papers. LNCS, vol. 15448, pp. 390–397. Springer, Cham (2025). https://doi.org/1...

work page doi:10.1007/978- 2025
[12]

In: Camara, O., et al

Merino-Caviedes, S., Nnadozie, E.C., Moya-Sáez, E., Martín-Fernández, M., Alberola-López, C.: CavityBASNet: cavity-focused biatrial automatic segmentation on LGE-MRI with augmented input channel and left-right myocardium splitting. In: Camara, O., et al. (eds.) Statistical Atlases and Computational Models of the Heart. Workshop, CMRxRecon and MBAS Challen...

work page doi:10.1007/978-3-031-87756-8_6 2025
[13]

et al.: SegBAW-Net: segmentation of bi-atria and wall network offering valuable insights into challenge data

Lefebvre, A.L. et al.: SegBAW-Net: segmentation of bi-atria and wall network offering valuable insights into challenge data. In: Camara, O., et al. (eds.) Statistical Atlases and Computational Models of the Heart. Workshop, CMRxRecon and MBAS Challenge Papers. LNCS, vol. 15448, pp. 345–356. Springer, Cham (2025). https://doi.org/10.1007/978-3-031-87756-8_34

work page doi:10.1007/978-3-031-87756-8_34 2025
[14]

arXiv preprint arXiv:2503.02892 (2025)

Gunawardhana, M., Trew, M.L., Sands, G.B., Zhao, J.: Segmenting bi-atrial structures using ResNext based framework . arXiv preprint arXiv:2503.02892 (2025). https://doi.org/10.48550/arXiv.2503.02892

work page doi:10.48550/arxiv.2503.02892 2025
[15]

In: Wallach, H.M., Larochelle, H., Beygelzimer, A., d'Alché-Buc, F., Fox, E.A., Garnett, R

Raghu, M., Zhang, C., Kleinberg, J., Bengio, S.: Transfusion: understanding transfer learning for medical imaging. In: Wallach, H.M., Larochelle, H., Beygelzimer, A., d'Alché-Buc, F., Fox, E.A., Garnett, R. (eds.) Advances in Neural Information Processing Systems 32, pp. 3342–3352. Curran Associates, Inc. (2019)

work page 2019
[16]

Adversarial discrim- inative domain adaptation,

Xie, S., Girshick, R., Dollar, P., Tu, Z., He, K.: Aggregated residual transformations for deep neural networks . In: 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , pp. 5987 –5995. IEEE, Honolulu, HI (2017). https://doi.org/10.1109/CVPR.2017.634

work page doi:10.1109/cvpr.2017.634 2017
[17]

et al.: Convolutional neural networks for medical image analysis: full training or fine tuning? IEEE Trans

Tajbakhsh, N. et al.: Convolutional neural networks for medical image analysis: full training or fine tuning? IEEE Trans. Med. Imaging. 35(5), 1299–1312 (2016). https://doi.org/10.1109/TMI.2016.2535302

work page doi:10.1109/tmi.2016.2535302 2016
[18]

et al.: Domain generalization in deep learning for contrast-enhanced imaging

Sendra-Balcells, C. et al.: Domain generalization in deep learning for contrast-enhanced imaging. Comput. Biol. Med. 149, 106052 (2022). https://doi.org/10.1016/j.compbiomed.2022.106052

work page doi:10.1016/j.compbiomed.2022.106052 2022
[19]

, year =

Ma, J. et al.: Loss odyssey in medical image segmentation. Med. Image Anal. 71, 102035 (2021). https://doi.org/10.1016/j.media.2021.102035

work page doi:10.1016/j.media.2021.102035 2021
[20]

Zhang, Z., Wang, Z., Wang, X., Wang, K., Yuan, Y., Li, Q.: A novel network with enhanced edge information for left atrium segmentation from LGE-MRI. Front. Physiol. 15, 1478347 (2024). https://doi.org/10.3389/fphys.2024.1478347

work page doi:10.3389/fphys.2024.1478347 2024

[1] [1]

Li, L., Zimmer, V.A., Schnabel, J.A., Zhuang, X.: Medical image analysis on left atrial LGE MRI for atrial fibrillation studies: a review. Med. Image Anal. 77, 102360 (2022). https://doi.org/10.1016/j.media.2022.102360

work page doi:10.1016/j.media.2022.102360 2022

[2] [2]

A global benchmark of algorithms for segmenting the left atrium from late gadolinium-enhanced cardiac magnetic resonance imaging,

Xiong, Z. et al.: A global benchmark of algorithms for segmenting the left atrium from late gadolinium-enhanced cardiac magnetic resonance imaging. Med. Image Anal. 67, 101832 (2021). https://doi.org/10.1016/j.media.2020.101832

work page doi:10.1016/j.media.2020.101832 2021

[3] [3]

Maceira, A.M., Cosín-Sales, J., Roughton, M., Prasad, S.K., Pennell, D.J.: Reference left atrial dimensions and volumes by steady state free precession cardiovascular magnetic resonance. J. Cardiovasc. Magn. Reson. 12, 65 (2010). https://doi.org/10.1186/1532- 429X-12-65

work page doi:10.1186/1532- 2010

[4] [4]

et al.: A robust computational framework for estimating 3D bi-atrial chamber wall thickness

Wang, Y. et al.: A robust computational framework for estimating 3D bi-atrial chamber wall thickness. Comput. Biol. Med. 114, 103444 (2019). https://doi.org/10.1016/j.compbiomed.2019.103444

work page doi:10.1016/j.compbiomed.2019.103444 2019

[5] [5]

Petitjean, C., Dacher, J.-N.: A review of segmentation methods in short axis cardiac MR images. Med. Image Anal. 15(2), 169– 184 (2011). https://doi.org/10.1016/j.media.2010.12.004

work page doi:10.1016/j.media.2010.12.004 2011

[6] [6]

et al.: Left atrial segmentation in a few seconds using fully convolutional network and transfer learning

Puybareau, É. et al.: Left atrial segmentation in a few seconds using fully convolutional network and transfer learning . In: Pop, M. et al. (eds.) Statistical Atlases and Computational Models of the Heart. Atrial Segmentation and LV Quantification Challenges. LNCS, vol. 11395, pp. 339–347. Springer International, Cham (2019). https://doi.org/10.1007/978-...

work page doi:10.1007/978-3-030-12029-0_37 2019

[7] [7]

Li, L., Zimmer, V.A., Schnabel, J.A., Zhuang, X.: AtrialJSQnet: a new framework for joint segmentation and quantification of left atrium and scars incorporating spatial and shape information. Med. Image Anal. 76, (2022). https://doi.org/10.1016/j.media.2021.102303

work page doi:10.1016/j.media.2021.102303 2022

[8] [8]

An Ensemble of 3D Residual Encoder UNet Models for Solving Multi - class Bi-atrial Segmentation Challenge,

Zolotarev, A., Johnson, K., Khan, A., Slabaugh, G., Roney, C.: An ensemble of 3D residual encoder UNet models for solving multi-class bi-atrial segmentation challenge. In: Camara, O., et al. (eds.) Statistical Atlases and Computational Models of the Heart. Workshop, CMRxRecon and MBAS Challenge Papers. LNCS, vol. 15448, pp. 209 –213. Springer, Cham (2025)...

work page doi:10.1007/978-3-031-87756-8_21 2025

[9] [9]

Evaluating Convolution, Attention, and Mamba Based U -Net Models for Multi- class Bi -Atrial Segmentation from LGE-MRI,

Thesing, C., Bueno-Orovio, A., Banerjee, A.: Evaluating convolution, attention, and Mamba based U-Net models for multi-class bi-atrial segmentation from LGE-MRI. In: Camara, O., et al. (eds.) Statistical Atlases and Computational Models of the Heart. Workshop, CMRxRecon and MBAS Challenge Papers. LNCS, vol. 15448, pp. 214 –225. Springer, Cham (2025). http...

work page doi:10.1007/978-3-031-87756-8_22 2025

[10] [10]

Multi-loss 3D Segmentation for Enhanced Bi- atrial Segmentation,

Almar-Munoz, E. et al.: Multi -loss 3D segmentation for enhanced bi-atrial segmentation. In: Camara, O., et al. (eds.) Statistical Atlases and Computational Models of the Heart. Workshop, CMRxRecon and MBAS Challenge Papers. LNCS, vol. 15448, pp. 236–244. Springer, Cham (2025). https://doi.org/10.1007/978-3-031-87756-8_24

work page doi:10.1007/978-3-031-87756-8_24 2025

[11] [11]

In: Proc

Zhang, Y., Cheng, Y., Qi, Y.: Exploring CNN and transformer architectures for multi -class bi -atrial segmentation from late gadolinium-enhanced MRI. In: Camara, O., et al. (eds.) Statistical Atlases and Computational Models of the Heart. Workshop, CMRxRecon and MBAS Challenge Papers. LNCS, vol. 15448, pp. 390–397. Springer, Cham (2025). https://doi.org/1...

work page doi:10.1007/978- 2025

[12] [12]

In: Camara, O., et al

Merino-Caviedes, S., Nnadozie, E.C., Moya-Sáez, E., Martín-Fernández, M., Alberola-López, C.: CavityBASNet: cavity-focused biatrial automatic segmentation on LGE-MRI with augmented input channel and left-right myocardium splitting. In: Camara, O., et al. (eds.) Statistical Atlases and Computational Models of the Heart. Workshop, CMRxRecon and MBAS Challen...

work page doi:10.1007/978-3-031-87756-8_6 2025

[13] [13]

et al.: SegBAW-Net: segmentation of bi-atria and wall network offering valuable insights into challenge data

Lefebvre, A.L. et al.: SegBAW-Net: segmentation of bi-atria and wall network offering valuable insights into challenge data. In: Camara, O., et al. (eds.) Statistical Atlases and Computational Models of the Heart. Workshop, CMRxRecon and MBAS Challenge Papers. LNCS, vol. 15448, pp. 345–356. Springer, Cham (2025). https://doi.org/10.1007/978-3-031-87756-8_34

work page doi:10.1007/978-3-031-87756-8_34 2025

[14] [14]

arXiv preprint arXiv:2503.02892 (2025)

Gunawardhana, M., Trew, M.L., Sands, G.B., Zhao, J.: Segmenting bi-atrial structures using ResNext based framework . arXiv preprint arXiv:2503.02892 (2025). https://doi.org/10.48550/arXiv.2503.02892

work page doi:10.48550/arxiv.2503.02892 2025

[15] [15]

In: Wallach, H.M., Larochelle, H., Beygelzimer, A., d'Alché-Buc, F., Fox, E.A., Garnett, R

Raghu, M., Zhang, C., Kleinberg, J., Bengio, S.: Transfusion: understanding transfer learning for medical imaging. In: Wallach, H.M., Larochelle, H., Beygelzimer, A., d'Alché-Buc, F., Fox, E.A., Garnett, R. (eds.) Advances in Neural Information Processing Systems 32, pp. 3342–3352. Curran Associates, Inc. (2019)

work page 2019

[16] [16]

Adversarial discrim- inative domain adaptation,

Xie, S., Girshick, R., Dollar, P., Tu, Z., He, K.: Aggregated residual transformations for deep neural networks . In: 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , pp. 5987 –5995. IEEE, Honolulu, HI (2017). https://doi.org/10.1109/CVPR.2017.634

work page doi:10.1109/cvpr.2017.634 2017

[17] [17]

et al.: Convolutional neural networks for medical image analysis: full training or fine tuning? IEEE Trans

Tajbakhsh, N. et al.: Convolutional neural networks for medical image analysis: full training or fine tuning? IEEE Trans. Med. Imaging. 35(5), 1299–1312 (2016). https://doi.org/10.1109/TMI.2016.2535302

work page doi:10.1109/tmi.2016.2535302 2016

[18] [18]

et al.: Domain generalization in deep learning for contrast-enhanced imaging

Sendra-Balcells, C. et al.: Domain generalization in deep learning for contrast-enhanced imaging. Comput. Biol. Med. 149, 106052 (2022). https://doi.org/10.1016/j.compbiomed.2022.106052

work page doi:10.1016/j.compbiomed.2022.106052 2022

[19] [19]

, year =

Ma, J. et al.: Loss odyssey in medical image segmentation. Med. Image Anal. 71, 102035 (2021). https://doi.org/10.1016/j.media.2021.102035

work page doi:10.1016/j.media.2021.102035 2021

[20] [20]

Zhang, Z., Wang, Z., Wang, X., Wang, K., Yuan, Y., Li, Q.: A novel network with enhanced edge information for left atrium segmentation from LGE-MRI. Front. Physiol. 15, 1478347 (2024). https://doi.org/10.3389/fphys.2024.1478347

work page doi:10.3389/fphys.2024.1478347 2024