pith. sign in

arxiv: 2604.17118 · v1 · submitted 2026-04-18 · 📡 eess.IV · cs.AI· cs.CV

A Two-Stage Deep Learning Framework for Segmentation of Ten Gastrointestinal Organs from Coronal MR Enterography

Ashiqur Rahman , Md. Abu Sayed , Md Sharjis Ibne Wadud , Md. Abu Asad Al-Hafiz , Adam Mushtak , Muhammad E. H. Chowdhury This is my paper

Pith reviewed 2026-05-10 06:19 UTC · model grok-4.3

classification 📡 eess.IV cs.AIcs.CV
keywords gastrointestinal organ segmentationMR enterographydeep learningtwo-stage frameworkinflammatory bowel diseaseU-Net variantsDice similarity coefficientmedical image analysis
0
0 comments X

The pith

A two-stage deep learning pipeline first locates broad regions then refines ten gastrointestinal organs in coronal MR enterography scans, reaching 88.99 percent mean Dice score.

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

The paper describes a dual-stage system that first uses a DenseNet201-UNet++ to create coarse masks and extract regions of interest from coronal T2-weighted images. A second model, DenseNet121-SelfONN-UNet, then segments each organ from targeted patches while applying class weighting and augmentation to handle imbalance. Tested on 3,195 slices from 114 IBD patients, the approach yields 88.99 percent mean Dice, 84.76 percent mean IoU, and 6.94 mm mean HD95, with largest gains for the appendix, cecum, sigmoid, and rectum. If the gains hold, the method could support automated tools that help diagnose and track inflammatory bowel disease. The authors note the extra compute cost but highlight the accuracy improvement over single-stage baselines.

Core claim

The authors claim that separating localization from organ-specific refinement in a coarse-to-fine pipeline overcomes low contrast and class imbalance in MRE images. The first stage produces usable ROIs for all ten structures; the second stage then delivers precise boundaries, lifting DSC by up to 23.62 percent for the cecum and 18.57 percent for the sigmoid. Overall metrics on the 114-patient public set exceed those of competing single-stage networks.

What carries the argument

The two-stage coarse-to-fine pipeline: DenseNet201-UNet++ for initial ROI extraction followed by DenseNet121-SelfONN-UNet for patch-wise refinement with class-specific weighting.

If this is right

  • The second stage produces measurable DSC gains for small and low-contrast organs that single-stage models miss.
  • Class weighting and patch-based training reduce the impact of severe imbalance, especially for the appendix.
  • The framework supplies anatomically detailed masks that could feed into downstream diagnostic or monitoring software for IBD.
  • Higher computational cost is accepted in exchange for the observed boundary accuracy.

Where Pith is reading between the lines

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

  • The same coarse-to-fine split could be tested on other abdominal MRI sequences or CT data to check transferability.
  • Reducing the second-stage model size while preserving the accuracy lift would make the pipeline more suitable for routine clinical workstations.
  • Combining the output masks with quantitative measures of organ wall thickness or inflammation could support automated IBD severity scoring.

Load-bearing premise

That the accuracy gains seen on this single 114-patient public dataset will hold for images from new patients, different scanners, or changed imaging protocols.

What would settle it

Running the trained models on an independent set of MRE scans acquired on different MRI machines or from a new patient population and checking whether the mean Dice remains above 85 percent would confirm or refute generalization.

Figures

Figures reproduced from arXiv: 2604.17118 by Adam Mushtak, Ashiqur Rahman, Md. Abu Asad Al-Hafiz, Md. Abu Sayed, Md Sharjis Ibne Wadud, Muhammad E. H. Chowdhury.

Figure 1
Figure 1. Figure 1: Overview of the anatomy of human digestive system [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Summary of the proposed framework. 3.1 Dataset Description This study employed a publicly available MRE dataset designed to support intestinal segmentation research in patients with IBD [29]. The dataset consists of 3,195 coronal T2-weighted HASTE slices obtained from 114 patients diagnosed with Crohn’s disease. Imaging was performed using 3.0-T Siemens Prisma and Vida scanners, following standardized bowe… view at source ↗
Figure 3
Figure 3. Figure 3: Architecture of initial stage multi-class segmentation model, DenseNet201-UNet++. [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Volumetric ROIs were extracted by stacking 2D predictions into 3D segmentation volumes. Tight bounding boxes with 40-pixel padding were computed for each organ to localize ROIs in the original MRE. These localized patches were used as input for second-stage organ-specific binary segmentation. Example shown for subject 72, slice 15. This volumetric approach was designed to address the limitations of convent… view at source ↗
Figure 5
Figure 5. Figure 5: Architecture of the proposed novel class-wise binary segmentation model. [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
read the original abstract

Accurate segmentation of gastrointestinal (GI) organs in magnetic resonance enterography (MRE) is critical for diagnosing inflammatory bowel disease (IBD). However, anatomical variability, class imbalance, and low tissue contrast hinder reliable automation. This study proposes a dual-stage deep learning framework for organ-specific segmentation of GI structures from coronal MRE images to address these challenges. A publicly available MRE dataset of 3,195 coronal T2-weighted HASTE slices from 114 IBD patients was used. Initially, a DenseNet201-UNet++ model generated coarse masks for ROI extraction. A DenseNet121-SelfONN-UNet model was then trained on organ-specific patches. Extensive data augmentation, normalization, five-fold cross-validation, and class-specific weighting were applied to mitigate severe class imbalance, particularly for the appendix. The initial stage achieved strong organ localization but underperformed for the appendix; class weighting improved its DSC from 6.76% to 85.76%. The second-stage DenseNet121-SelfONN-UNet significantly enhanced segmentation across all GI structures, with notable DSC gains (cecum +23.62%, sigmoid +18.57%, rectum +17.99%, small intestine +16.06%). Overall, the framework achieved mDSC of 88.99%, mIoU of 84.76%, and mHD95 of 6.94 mm, outperforming all baselines. This framework demonstrates the effectiveness of a coarse-to-fine, organ-aware segmentation strategy for intestinal MRE. Despite higher computational cost, it shows strong potential for clinical translation and enables anatomically informed diagnostic tools in gastroenterology.

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 / 2 minor

Summary. The manuscript describes a two-stage deep learning framework for the segmentation of ten gastrointestinal organs in coronal MR enterography (MRE) images. The first stage employs a DenseNet201-UNet++ model to produce coarse segmentation masks, which are used to extract organ-specific regions of interest (ROIs). The second stage then applies a DenseNet121-SelfONN-UNet model trained on these patches for refined segmentation. The approach is evaluated on a public dataset comprising 3,195 coronal T2-weighted HASTE slices from 114 IBD patients, utilizing five-fold cross-validation, data augmentation, normalization, and class-specific weighting to address class imbalance. The framework reports an overall mean Dice Similarity Coefficient (mDSC) of 88.99%, mean IoU of 84.76%, and mean HD95 of 6.94 mm, with notable improvements in the second stage for organs such as the cecum (+23.62% DSC), sigmoid (+18.57%), rectum (+17.99%), and small intestine (+16.06%), outperforming baseline models.

Significance. Should the reported performance metrics prove robust upon clarification of the evaluation protocol, this study makes a meaningful contribution to automated analysis of MRE for inflammatory bowel disease by showing how a coarse-to-fine, organ-aware strategy can mitigate challenges like anatomical variability and severe class imbalance (e.g., improving appendix DSC from 6.76% to 85.76% with weighting). The explicit use of five-fold cross-validation, augmentation, and class weighting are positive aspects that enhance reproducibility. The work has potential for clinical translation in gastroenterology, though external validation on diverse datasets would strengthen the claims of generalizability.

major comments (1)
  1. [Methods and Experiments (cross-validation procedure)] In the Methods section describing the two-stage pipeline and the Experiments section on five-fold cross-validation, it is not specified whether the first-stage DenseNet201-UNet++ model is retrained independently within each fold (i.e., nested CV) or if a single model trained on the full 3,195-slice dataset is used to generate coarse masks and organ-specific patches for the second stage. The latter case would allow test-set information to leak into the second-stage training via the first-stage predictions, rendering the central performance claims (mDSC 88.99%, mIoU 84.76%, and per-organ DSC gains such as +23.62% for cecum) unreliable.
minor comments (2)
  1. [Abstract] The abstract states that the framework 'outperforms all baselines' but does not name the specific baseline models (e.g., single-stage U-Net variants or other DenseNet configurations); adding this detail would allow readers to better contextualize the reported gains.
  2. [Abstract] The abstract refers to segmentation of 'Ten Gastrointestinal Organs' without listing them; explicitly naming the organs (appendix, cecum, sigmoid, rectum, small intestine, etc.) in the abstract would improve immediate clarity.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their careful review and for identifying an important ambiguity in our description of the evaluation protocol. We address the major comment below and will revise the manuscript accordingly to improve clarity and reproducibility.

read point-by-point responses

  1. Referee: [Methods and Experiments (cross-validation procedure)] In the Methods section describing the two-stage pipeline and the Experiments section on five-fold cross-validation, it is not specified whether the first-stage DenseNet201-UNet++ model is retrained independently within each fold (i.e., nested CV) or if a single model trained on the full 3,195-slice dataset is used to generate coarse masks and organ-specific patches for the second stage. The latter case would allow test-set information to leak into the second-stage training via the first-stage predictions, rendering the central performance claims (mDSC 88.99%, mIoU 84.76%, and per-organ DSC gains such as +23.62% for cecum) unreliable.

    Authors: We sincerely thank the referee for highlighting this critical detail. We confirm that a nested cross-validation procedure was used: within each of the five folds, the DenseNet201-UNet++ model was trained exclusively on the training subset of that fold (using the validation subset for early stopping and hyperparameter tuning), and the trained model was applied only to the held-out test subset of the same fold to produce coarse masks and extract organ-specific ROIs. The second-stage DenseNet121-SelfONN-UNet was then trained and evaluated solely on the patches derived from the training data of that fold. This ensures complete separation and prevents any test-set leakage. We apologize for the lack of explicit description in the original manuscript. In the revised version, we will expand the Methods and Experiments sections to detail the nested CV protocol, specify that partitioning was performed at the patient level, and include a schematic of the fold-wise process. revision: yes

Circularity Check

0 steps flagged

No significant circularity in empirical two-stage segmentation framework

full rationale

The paper describes a standard empirical deep learning pipeline: a public dataset of 3,195 slices is split via five-fold cross-validation, a first-stage DenseNet201-UNet++ produces coarse masks for patch extraction, and a second-stage DenseNet121-SelfONN-UNet is trained on those patches to report mDSC, mIoU, and mHD95 on held-out folds. No equations, uniqueness theorems, or ansatzes are invoked; performance numbers are direct outputs of training and evaluation rather than quantities that reduce to fitted inputs by construction. The two-stage design is a conventional coarse-to-fine strategy with no self-referential definitions or load-bearing self-citations. While the exact nesting of the first-stage model within each CV fold is not quoted in the provided text, this is a methodological detail rather than a circular reduction of the claimed metrics to the inputs themselves. The derivation chain is therefore self-contained and externally falsifiable on the public dataset.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on standard deep learning assumptions about data distribution and generalization plus the representativeness of the 114-patient MRE dataset; no new entities or ad-hoc constants beyond typical model hyperparameters.

free parameters (1)
  • class-specific weights
    Introduced to boost appendix segmentation from 6.76% to 85.76% DSC; chosen to counter severe imbalance.
axioms (1)
  • domain assumption The 3,195-slice dataset from 114 IBD patients is sufficiently representative for clinical generalization
    Invoked implicitly when claiming potential for clinical translation.

pith-pipeline@v0.9.0 · 5642 in / 1185 out tokens · 43459 ms · 2026-05-10T06:19:21.778809+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

38 extracted references · 38 canonical work pages

  1. [1]

    Large Association of GI Tract Microbial Community with Immune and Nervous Systems,

    A. Kazempour and A. Kazempour, “Large Association of GI Tract Microbial Community with Immune and Nervous Systems,” Immunology of the GI Tract - Recent Advances, Dec. 2022, doi: 10.5772/INTECHOPEN.104120

    work page doi:10.5772/intechopen.104120 2022

  2. [2]

    Pathophysiologic Role of Neurotransmitters in Digestive Diseases,

    X. Yang et al., “Pathophysiologic Role of Neurotransmitters in Digestive Diseases,” Front Physiol, vol. 12, p. 567650, Jun. 2021, doi: 10.3389/FPHYS.2021.567650/XML/NLM

    work page doi:10.3389/fphys.2021.567650/xml/nlm 2021

  3. [3]

    The enteric nervous system,

    K. A. Sharkey and G. M. Mawe, “The enteric nervous system,” Physiol Rev, vol. 103, no. 2, pp. 1487–1564, Apr. 2023, doi: 10.1152/PHYSREV.00018.2022/ASSET/IMAGES/LARGE/PHYSREV.00018.2022_F010.JPEG

    work page doi:10.1152/physrev.00018.2022/asset/images/large/physrev.00018.2022_f010.jpeg 2023

  4. [4]

    The Relationship Between Intestinal Microbiota and the Central Nervous System in Normal Gastrointestinal Function and Disease,

    S. M. Collins and P. Bercik, “The Relationship Between Intestinal Microbiota and the Central Nervous System in Normal Gastrointestinal Function and Disease,” Gastroenterology, vol. 136, no. 6, pp. 2003– 2014, May 2009, doi: 10.1053/J.GASTRO.2009.01.075

    work page doi:10.1053/j.gastro.2009.01.075 2003

  5. [5]

    Physiology, Gastrointestinal,

    I. Ogobuiro, J. Gonzales, K. R. Shumway, and F. Tuma, “Physiology, Gastrointestinal,” StatPearls, Apr. 2023, Accessed: Jul. 03, 2025. [Online]. Available: https://www.ncbi.nlm.nih.gov/books/NBK537103/

    work page 2023

  6. [6]

    illustration of Healthcare and Medical education drawing chart of Human Digestive System for Science Biology study 2803159 Vector Art at Vecteezy

    “illustration of Healthcare and Medical education drawing chart of Human Digestive System for Science Biology study 2803159 Vector Art at Vecteezy.” Accessed: Jul. 04, 2025. [Online]. Available: https://www.vecteezy.com/vector-art/2803159-illustration-of-healthcare-and-medical-education-drawing- chart-of-human-digestive-system-for-science-biology-study

    work page arXiv 2025

  7. [7]

    Epidemiology and clinical course of Crohn’s disease: Results from observational studies,

    Ø. Hovde and B. A. Moum, “Epidemiology and clinical course of Crohn’s disease: Results from observational studies,” World J Gastroenterol, vol. 18, no. 15, pp. 1723–1731, 2012, doi: 10.3748/WJG.V18.I15.1723

    work page doi:10.3748/wjg.v18.i15.1723 2012

  8. [8]

    Diagnostic Procedures for Inflammatory Bowel Disease: Laboratory, Endoscopy, Pathology, Imaging, and Beyond,

    S. M. Hong and D. H. Baek, “Diagnostic Procedures for Inflammatory Bowel Disease: Laboratory, Endoscopy, Pathology, Imaging, and Beyond,” Diagnostics 2024, Vol. 14, Page 1384, vol. 14, no. 13, p. 1384, Jun. 2024, doi: 10.3390/DIAGNOSTICS14131384

    work page doi:10.3390/diagnostics14131384 2024

  9. [9]

    Crohn’s disease: factors associated with exposure to high levels of diagnostic radiation,

    A. N. Desmond et al., “Crohn’s disease: factors associated with exposure to high levels of diagnostic radiation,” Gut, vol. 57, no. 11, pp. 1524–1529, Nov. 2008, doi: 10.1136/GUT.2008.151415

    work page doi:10.1136/gut.2008.151415 2008

  10. [10]

    Imaging in inflammatory bowel disease: current and future perspectives,

    N. Shaban et al., “Imaging in inflammatory bowel disease: current and future perspectives,” Frontline Gastroenterol, vol. 13, no. e1, pp. e28–e34, Aug. 2022, doi: 10.1136/FLGASTRO-2022-102117

    work page doi:10.1136/flgastro-2022-102117 2022

  11. [11]

    Magnetic resonance for assessment of disease activity and severity in ileocolonic Crohn’s disease,

    J. Rimola et al., “Magnetic resonance for assessment of disease activity and severity in ileocolonic Crohn’s disease,” Gut, vol. 58, no. 8, pp. 1113–1120, Aug. 2009, doi: 10.1136/GUT.2008.167957

    work page doi:10.1136/gut.2008.167957 2009

  12. [12]

    Endoscopic Skipping of the Distal Terminal Ileum in Crohn’s Disease Can Lead to Negative Results From Ileocolonoscopy,

    S. Samuel et al., “Endoscopic Skipping of the Distal Terminal Ileum in Crohn’s Disease Can Lead to Negative Results From Ileocolonoscopy,” Clinical Gastroenterology and Hepatology, vol. 10, no. 11, pp. 1253–1259, Nov. 2012, doi: 10.1016/J.CGH.2012.03.026

    work page doi:10.1016/j.cgh.2012.03.026 2012

  13. [13]

    Gastrointestinal diseases segmentation and classification based on duo-deep architectures,

    M. A. Khan et al., “Gastrointestinal diseases segmentation and classification based on duo-deep architectures,” Pattern Recognit Lett, vol. 131, pp. 193–204, Mar. 2020, doi: 10.1016/J.PATREC.2019.12.024

    work page doi:10.1016/j.patrec.2019.12.024 2020

  14. [14]

    MR imaging of the small bowel in Crohn’s disease,

    H. Siddiki and J. Fidler, “MR imaging of the small bowel in Crohn’s disease,” Eur J Radiol, vol. 69, no. 3, pp. 409–417, Mar. 2009, doi: 10.1016/J.EJRAD.2008.11.013

    work page doi:10.1016/j.ejrad.2008.11.013 2009

  15. [15]

    Inflammatory Bowel Disease Diagnosed with US, MR, Scintigraphy, and CT: Meta-analysis of Prospective Studies1,

    K. Horsthuis, S. Bipat, R. J. Bennink, and J. Stoker, “Inflammatory Bowel Disease Diagnosed with US, MR, Scintigraphy, and CT: Meta-analysis of Prospective Studies1,” pubs.rsna.orgK Horsthuis, S Bipat, RJ Bennink, J StokerRadiology, 2008•pubs.rsna.org, vol. 247, no. 1, pp. 64–79, Apr. 2008, doi: 10.1148/RADIOL.2471070611

    work page doi:10.1148/radiol.2471070611 2008

  16. [16]

    Deep convolutional neural networks for multi-modality isointense infant brain image segmentation,

    W. Zhang et al., “Deep convolutional neural networks for multi-modality isointense infant brain image segmentation,” Neuroimage, vol. 108, pp. 214–224, Mar. 2015, doi: 10.1016/J.NEUROIMAGE.2014.12.061

    work page doi:10.1016/j.neuroimage.2014.12.061 2015

  17. [17]

    MAPGI: Accurate identification of anatomical landmarks and diseased tissue in gastrointestinal tract using deep learning,

    T. Cogan, M. Cogan, and L. Tamil, “MAPGI: Accurate identification of anatomical landmarks and diseased tissue in gastrointestinal tract using deep learning,” Comput Biol Med, vol. 111, p. 103351, Aug. 2019, doi: 10.1016/J.COMPBIOMED.2019.103351

    work page doi:10.1016/j.compbiomed.2019.103351 2019

  18. [18]

    Automatic Multi-Organ Segmentation on Abdominal CT with Dense V-Networks,

    E. Gibson et al., “Automatic Multi-Organ Segmentation on Abdominal CT with Dense V-Networks,” IEEE Trans Med Imaging, vol. 37, no. 8, pp. 1822–1834, Aug. 2018, doi: 10.1109/TMI.2018.2806309

    work page doi:10.1109/tmi.2018.2806309 2018

  19. [19]

    Multi-scale context-guided deep network for automated lesion segmentation with endoscopy images of gastrointestinal tract,

    S. Wang, Y. Cong, H. Zhu, X. Chen, … L. Q.-I. J. of, and undefined 2020, “Multi-scale context-guided deep network for automated lesion segmentation with endoscopy images of gastrointestinal tract,” ieeexplore.ieee.orgS Wang, Y Cong, H Zhu, X Chen, L Qu, H Fan, Q Zhang, M LiuIEEE Journal of Biomedical and Health Informatics, 2020•ieeexplore.ieee.org, Acces...

    work page arXiv 2020

  20. [20]

    Deep learning for automatic bowel-obstruction identification on abdominal CT,

    Q. Vanderbecq et al., “Deep learning for automatic bowel-obstruction identification on abdominal CT,” SpringerQ Vanderbecq, M Gelard, JC Pesquet, M Wagner, L Arrive, M Zins, E ChouzenouxEuropean Radiology, 2024•Springer, vol. 34, no. 9, pp. 5842–5853, Sep. 2024, doi: 10.1007/S00330-024-10657-Z

    work page doi:10.1007/s00330-024-10657-z 2024

  21. [21]

    The application of deep learning in abdominal trauma diagnosis by CT imaging,

    X. Shen et al., “The application of deep learning in abdominal trauma diagnosis by CT imaging,” SpringerX Shen, Y Zhou, X Shi, S Zhang, S Ding, L Ni, X Dou, L ChenWorld Journal of Emergency Surgery, 2024•Springer, vol. 19, no. 1, Dec. 2024, doi: 10.1186/S13017-024-00546-7

    work page doi:10.1186/s13017-024-00546-7 2024

  22. [22]

    Semi-automatic sigmoid colon segmentation in CT for radiation therapy treatment planning via an iterative 2.5-D deep learning approach,

    Y. Gonzalez et al., “Semi-automatic sigmoid colon segmentation in CT for radiation therapy treatment planning via an iterative 2.5-D deep learning approach,” Med Image Anal, vol. 68, Feb. 2021, doi: 10.1016/j.media.2020.101896

    work page doi:10.1016/j.media.2020.101896 2021

  23. [23]

    Y. Lamash et al., “Curved planar reformatting and convolutional neural network-based segmentation of the small bowel for visualization and quantitative assessment of pediatric Crohn’s disease from MRI,” Journal of Magnetic Resonance Imaging, vol. 49, no. 6, pp. 1565–1576, Jun. 2019, doi: 10.1002/jmri.26330

    work page doi:10.1002/jmri.26330 2019

  24. [24]

    Garrucho, K

    L. D. van Harten, C. S. de Jonge, K. J. Beek, J. Stoker, and I. Išgum, “Untangling and segmenting the small intestine in 3D cine-MRI using deep learning,” Med Image Anal, vol. 78, May 2022, doi: 10.1016/j.media.2022.102386

    work page doi:10.1016/j.media.2022.102386 2022

  25. [25]

    Magnetic resonance imaging of the gastrointestinal tract shows reduced small bowel motility and altered chyme in cystic fibrosis compared to controls,

    N. S. Dellschaft et al., “Magnetic resonance imaging of the gastrointestinal tract shows reduced small bowel motility and altered chyme in cystic fibrosis compared to controls,” Journal of Cystic Fibrosis, vol. 21, no. 3, pp. 502–505, May 2022, doi: 10.1016/j.jcf.2021.12.007

    work page doi:10.1016/j.jcf.2021.12.007 2022

  26. [26]

    Automatic colon segmentation on T1-FS MR images,

    B. Orellana, I. Navazo, P. Brunet, E. Monclús, Á. Bendezú, and F. Azpiroz, “Automatic colon segmentation on T1-FS MR images,” Computerized Medical Imaging and Graphics, vol. 123, Jul. 2025, doi: 10.1016/j.compmedimag.2025.102528

    work page doi:10.1016/j.compmedimag.2025.102528 2025

  27. [27]

    Automatic Contour Refinement for Deep Learning Auto-segmentation of Complex Organs in MRI-guided Adaptive Radiation Therapy,

    J. Ding, Y. Zhang, A. Amjad, J. Xu, D. Thill, and X. A. Li, “Automatic Contour Refinement for Deep Learning Auto-segmentation of Complex Organs in MRI-guided Adaptive Radiation Therapy,” Adv Radiat Oncol, vol. 7, no. 5, Sep. 2022, doi: 10.1016/j.adro.2022.100968

    work page doi:10.1016/j.adro.2022.100968 2022

  28. [28]

    Deep learning in magnetic resonance enterography for Crohn’s disease assessment: a systematic review,

    O. Brem, D. Elisha, E. Konen, M. Amitai, and E. Klang, “Deep learning in magnetic resonance enterography for Crohn’s disease assessment: a systematic review,” Sep. 01, 2024, Springer. doi: 10.1007/s00261-024-04326-4

    work page doi:10.1007/s00261-024-04326-4 2024

  29. [29]

    A comprehensive dataset of magnetic resonance enterography images with intestinal segment annotations,

    Z. Zhong et al., “A comprehensive dataset of magnetic resonance enterography images with intestinal segment annotations,” Sci Data, vol. 12, no. 1, p. 425, Dec. 2025, doi: 10.1038/S41597-025-04760- Z;SUBJMETA=1046,1503,257,639,692,699,705;KWRD=INFLAMMATORY+BOWEL+DISEASE,SCIE NTIFIC+DATA

    work page doi:10.1038/s41597-025-04760- 2025

  30. [30]

    , author Siddiquee, M.M.R

    Z. Zhou, M. M. Rahman Siddiquee, N. Tajbakhsh, and J. Liang, “UNet++: A Nested U-Net Architecture for Medical Image Segmentation,” Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 11045 LNCS, pp. 3–11, Jul. 2018, doi: 10.1007/978-3-030-00889-5_1

    work page doi:10.1007/978-3-030-00889-5_1 2018

  31. [31]

    In: 2017 IEEE Con- ference on Computer Vision and Pattern Recognition (CVPR)

    G. Huang, Z. Liu, L. Van Der Maaten, and K. Q. Weinberger, “Densely Connected Convolutional Networks,” Proceedings - 30th IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2017, vol. 2017-January, pp. 2261–2269, Aug. 2016, doi: 10.1109/CVPR.2017.243

    work page doi:10.1109/cvpr.2017.243 2017

  32. [32]

    Self-organized Operational Neural Networks with Generative Neurons,

    S. Kiranyaz, J. Malik, H. Ben Abdallah, T. Ince, A. Iosifidis, and M. Gabbouj, “Self-organized Operational Neural Networks with Generative Neurons,” Neural Networks, vol. 140, pp. 294–308, Aug. 2021, doi: 10.1016/J.NEUNET.2021.02.028

    work page doi:10.1016/j.neunet.2021.02.028 2021

  33. [33]

    Deep learning-driven segmentation of ischemic stroke lesions using multi-channel MRI,

    A. Rahman et al., “Deep learning-driven segmentation of ischemic stroke lesions using multi-channel MRI,” Biomed Signal Process Control, vol. 105, Jul. 2025, doi: 10.1016/j.bspc.2025.107676

    work page doi:10.1016/j.bspc.2025.107676 2025

  34. [34]

    H., & Yamamoto, K

    H. Phan, K. Yamamoto, T. H. Phan, and K. Yamamoto, “Resolving Class Imbalance in Object Detection with Weighted Cross Entropy Losses,” Jun. 2020, Accessed: Jul. 03, 2025. [Online]. Available: https://arxiv.org/pdf/2006.01413

    work page arXiv 2020

  35. [35]

    Fuller, A., Millard, K., Green, J., 2023

    S. Jadon, “A survey of loss functions for semantic segmentation,” 2020 IEEE Conference on Computational Intelligence in Bioinformatics and Computational Biology, CIBCB 2020, Oct. 2020, doi: 10.1109/CIBCB48159.2020.9277638

    work page doi:10.1109/cibcb48159.2020.9277638 2020

  36. [36]

    Metrics for evaluating 3D medical image segmentation: Analysis, selection, and tool,

    A. A. Taha and A. Hanbury, “Metrics for evaluating 3D medical image segmentation: Analysis, selection, and tool,” BMC Med Imaging, vol. 15, no. 1, pp. 1–28, Aug. 2015, doi: 10.1186/S12880-015-0068- X/TABLES/5

    work page doi:10.1186/s12880-015-0068- 2015

  37. [37]

    Towards a guideline for evaluation metrics in medical image segmentation,

    D. Müller, I. Soto-Rey, and F. Kramer, “Towards a guideline for evaluation metrics in medical image segmentation,” BMC Res Notes, vol. 15, no. 1, pp. 1–8, Dec. 2022, doi: 10.1186/S13104-022-06096- Y/FIGURES/2

    work page doi:10.1186/s13104-022-06096- 2022

  38. [38]

    Establishing a machine learning model based on dual-energy CT enterography to evaluate Crohn’s disease activity,

    J. Li et al., “Establishing a machine learning model based on dual-energy CT enterography to evaluate Crohn’s disease activity,” Insights Imaging, vol. 15, no. 1, Dec. 2024, doi: 10.1186/s13244-024-01703-x

    work page doi:10.1186/s13244-024-01703-x 2024