pith. sign in

arxiv: 2604.26251 · v1 · submitted 2026-04-29 · 💻 cs.CV · cs.AI· cs.LG

Multi-Stage Bi-Atrial Segmentation Framework from 3D Late Gadolinium-Enhanced MRI using V-Net Family Models

Hao Wen , Jingsu Kang This is my paper

Pith reviewed 2026-05-07 14:01 UTC · model grok-4.3

classification 💻 cs.CV cs.AIcs.LG
keywords bi-atrial segmentation3D LGE MRIV-Netmulti-stage frameworkMCLAHEasymmetric losscardiac MRI
0
0 comments X

The pith

A multi-stage V-Net pipeline segments bi-atrial structures from 3D LGE MRI after MCLAHE preprocessing.

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

The paper describes a multi-stage framework for multi-class bi-atrial segmentation from 3D late gadolinium-enhanced MRI of the heart. It applies multidimensional contrast limited adaptive histogram equalization for preprocessing, performs coarse segmentation on down-sampled enhanced volumes with a V-Net family model, and refines the result with a second V-Net on the extracted coarse region. Asymmetric loss trains the models to handle the task. A sympathetic reader would care because precise automated delineation of the left and right atria supports diagnosis of atrial fibrillation and fibrosis without exhaustive manual tracing.

Core claim

We report our multi-stage framework designed for the problem of multi-class bi-atrial segmentation from 3D late gadolinium-enhanced (LGE) MRI of the human heart. The pipeline consists of a preprocessing step using multidimensional contrast limited adaptive histogram equalization (MCLAHE); coarse region segmentation from MCLAHE-enhanced and down-sampled MRI using a V-Net family model; and fine segmentation from the coarse region using another V-Net model. Asymmetric loss is adopted to optimize the model weights.

What carries the argument

The multi-stage pipeline that uses MCLAHE for contrast enhancement, a first V-Net for coarse region detection on downsampled data, and a second V-Net for fine boundary segmentation inside the coarse mask.

Load-bearing premise

The multi-stage V-Net pipeline will produce accurate and generalizable segmentations on real clinical data even though no quantitative metrics or comparisons are supplied.

What would settle it

Running the pipeline on an independent held-out set of 3D LGE MRI volumes and measuring Dice scores, Hausdorff distances, and surface errors against manual ground truth to see whether performance exceeds single-stage baselines.

Figures

Figures reproduced from arXiv: 2604.26251 by Hao Wen, Jingsu Kang.

Figure 1
Figure 1. Figure 1: The enhanced MRIs are rescaled by zero-padding and cropping so that view at source ↗
Figure 2
Figure 2. Figure 2: Loss curves of the models on the training data. view at source ↗
read the original abstract

We report our multi-stage framework designed for the problem of multi-class bi-atrial segmentation from 3D late gadolinium-enhanced (LGE) MRI of the human heart. The pipeline consists of a preprocessing step using multidimensional contrast limited adaptive histogram equalization (MCLAHE); coarse region segmentation from MCLAHE-enhanced and down-sampled MRI using a V-Net family model; and fine segmentation from the coarse region using another V-Net model. Asymmetric loss is adopted to optimize the model weights.

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.

Referee Report

1 major / 0 minor

Summary. The manuscript describes a multi-stage framework for multi-class bi-atrial segmentation from 3D late gadolinium-enhanced (LGE) MRI. The pipeline consists of MCLAHE preprocessing, coarse segmentation of down-sampled volumes via a V-Net family model, fine segmentation of the cropped region via a second V-Net model, and training with an asymmetric loss function.

Significance. If empirically validated, the coarse-to-fine V-Net pipeline with MCLAHE could offer a practical solution for accurate bi-atrial segmentation in cardiac LGE-MRI, addressing challenges of small structures and class imbalance in 3D volumes. The modular design and asymmetric loss are reasonable engineering choices for this task.

major comments (1)
  1. The manuscript contains no Results section, no tables or figures reporting quantitative metrics (Dice, IoU, Hausdorff distance, etc.), no ablation studies, and no comparisons against single-stage baselines or prior atrial segmentation methods. This is load-bearing for the central claim that the multi-stage framework solves the bi-atrial segmentation problem, as the abstract and methods description alone provide no evidence of accuracy or generalizability on clinical data.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We agree that empirical validation through quantitative results is essential to support the claims of the multi-stage framework, and we will revise the manuscript to address this gap.

read point-by-point responses

  1. Referee: The manuscript contains no Results section, no tables or figures reporting quantitative metrics (Dice, IoU, Hausdorff distance, etc.), no ablation studies, and no comparisons against single-stage baselines or prior atrial segmentation methods. This is load-bearing for the central claim that the multi-stage framework solves the bi-atrial segmentation problem, as the abstract and methods description alone provide no evidence of accuracy or generalizability on clinical data.

    Authors: We acknowledge that the current manuscript version is primarily methodological and lacks a dedicated Results section with supporting quantitative evidence. This is a valid and important observation. In the revised manuscript, we will add a comprehensive Results section that includes tables and figures reporting Dice scores, IoU, Hausdorff distances, and other relevant metrics evaluated on clinical 3D LGE-MRI datasets. We will also incorporate ablation studies comparing the full multi-stage pipeline (with MCLAHE and asymmetric loss) against single-stage V-Net baselines, as well as comparisons to existing bi-atrial segmentation approaches from the literature. These additions will directly demonstrate the accuracy and generalizability of the proposed framework. revision: yes

Circularity Check

0 steps flagged

No circularity; purely descriptive methods paper with no equations or derivation chain

full rationale

The manuscript describes a multi-stage V-Net pipeline (MCLAHE preprocessing, coarse downsampled segmentation, fine cropped segmentation, asymmetric loss) but contains no equations, no first-principles derivations, no fitted parameters presented as predictions, and no self-citation chains that reduce claims to inputs. All load-bearing steps are standard architectural choices and preprocessing operations whose correctness is external to the paper; no step reduces by construction to another. The lack of quantitative results is a separate empirical-support issue, not a circularity problem.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The work is purely applied and contains no free parameters, axioms, or invented entities; it relies on standard deep-learning assumptions such as the suitability of V-Net for volumetric segmentation.

pith-pipeline@v0.9.0 · 5380 in / 1058 out tokens · 49405 ms · 2026-05-07T14:01:06.579550+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

11 extracted references · 11 canonical work pages

  1. [1]

    CodaLab Competitions: An Open Source Platform to Organize Scientific Challenges,

    A. Pavao, I. Guyon, A.-C. Letournel, D.-T. Tran, X. Baro, H. J. Escalante, S. Escalera, T. Thomas, and Z. Xu, “CodaLab Competitions: An Open Source Platform to Organize Scientific Challenges,”Journal of Machine Learning Research, vol. 24, no. 198, pp. 1–6, 2023. [Online]. Available: http://jmlr.org/papers/v24/21-1436.html

    work page 2023

  2. [2]

    A Global Benchmark of Algorithms for Segmenting the Left Atrium from Late Gadolinium- Enhanced Cardiac Magnetic Resonance Imaging,

    Z. Xiong, Q. Xia, Z. Hu, N. Huang, C. Bian, Y. Zheng, S. Vesal, N. Ravikumar, A. Maier, X. Yang, P.-A. Heng, D. Ni, C. Li, Q. Tong, W. Si, E. Puybareau, Y. Khoudli, T. Géraud, C. Chen, W. Bai, D. Rueckert, L. Xu, X. Zhuang, X. Luo, S. Jia, M. Sermesant, Y. Liu, K. Wang, D. Borra, A. Masci, C. Corsi, C. de Vente, M. Veta, R. Karim, C. J. Preetha, S. Engelh...

    work page 2021

  3. [3]

    Multidimen- sional Contrast Limited Adaptive Histogram Equalization,

    V. Stimper, S. Bauer, R. Ernstorfer, B. Scholkopf, and R. P. Xian, “Multidimen- sional Contrast Limited Adaptive Histogram Equalization,”IEEE Access, vol. 7, p. 165437–165447, 2019

    work page 2019

  4. [4]

    V-Net: Fully Convolutional Neural Net- works for Volumetric Medical Image Segmentation,

    F. Milletari, N. Navab, and S.-A. Ahmadi, “V-Net: Fully Convolutional Neural Net- works for Volumetric Medical Image Segmentation,” in2016 Fourth International Conference on 3D Vision (3DV). IEEE, 10 2016, pp. 565–571

    work page 2016

  5. [5]

    UNet++: A Nested U-Net Architecture for Medical Image Segmentation,

    Z. Zhou, M. M. Rahman Siddiquee, N. Tajbakhsh, and J. Liang, “UNet++: A Nested U-Net Architecture for Medical Image Segmentation,” inDeep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support, 6 Hao Wen and Jingsu Kang D. Stoyanov, Z. Taylor, G. Carneiro, T. Syeda-Mahmood, A. Martel, L. Maier- Hein, J. M. R. Tavares, A. ...

    work page 2018

  6. [6]

    Asymmetric Loss for Multi-Label Classification,

    T. Ridnik, E. Ben-Baruch, N. Zamir, A. Noy, I. Friedman, M. Protter, and L. Zelnik-Manor, “Asymmetric Loss for Multi-Label Classification,” inProceed- ings of the IEEE/CVF International Conference on Computer Vision. IEEE, 10 2021, pp. 82–91

    work page 2021

  7. [7]

    On the Convergence of Adam and Beyond,

    S. J. Reddi, S. Kale, and S. Kumar, “On the Convergence of Adam and Beyond,” in6th International Conference on Learning Representations (ICLR). OpenRe- view.net, 5 2018. [Online]. Available: https://openreview.net/forum?id=ryQu7f- RZ

    work page 2018

  8. [8]

    Super-Convergence: Very Fast Training of Neural Net- works Using Large Learning Rates,

    L. N. Smith and N. Topin, “Super-Convergence: Very Fast Training of Neural Net- works Using Large Learning Rates,” inArtificial Intelligence and Machine Learning for Multi-Domain Operations Applications, T. Pham, Ed., vol. 11006, International Society for Optics and Photonics. SPIE, 4 2019, p. 1100612

    work page 2019

  9. [9]

    UNETR: Transformers for 3D Medical Image Segmentation,

    A. Hatamizadeh, Y. Tang, V. Nath, D. Yang, A. Myronenko, B. Landman, H. R. Roth, and D. Xu, “UNETR: Transformers for 3D Medical Image Segmentation,” inProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 1 2022, pp. 574–584

    work page 2022

  10. [10]

    Segment Anything in Medical Images,

    J. Ma, Y. He, F. Li, L. Han, C. You, and B. Wang, “Segment Anything in Medical Images,”Nature Communications, vol. 15, no. 1, 1 2024

    work page 2024

  11. [11]

    Loss Odyssey in Medical Image Segmentation,

    J. Ma, J. Chen, M. Ng, R. Huang, Y. Li, C. Li, X. Yang, and A. L. Martel, “Loss Odyssey in Medical Image Segmentation,”Medical Image Analysis, vol. 71, p. 102035, 7 2021

    work page 2021