pith. machine review for the scientific record. sign in

arxiv: 2604.10702 · v3 · submitted 2026-04-12 · 💻 cs.CV · cs.AI

Recognition: unknown

Architecture-Agnostic Modality-Isolated Gated Fusion for Robust Multi-Modal Prostate MRI Segmentation

Yongbo Shu , Wenzhao Xie , Shanhu Yao , Zirui Xin , Luo Lei , Kewen Chen , Aijing Luo
Authors on Pith no claims yet

Pith reviewed 2026-05-10 15:54 UTC · model grok-4.3

classification 💻 cs.CV cs.AI
keywords prostate MRI segmentationmulti-modal fusionmissing modalitiesgated fusionmodality dropoutrobust medical imagingdiffusion-weighted MRI
0
0 comments X

The pith

Modality-isolated gated fusion with dropout training makes multi-modal prostate MRI segmentation more robust to missing or degraded diffusion sequences.

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

The paper proposes a fusion approach that keeps T2-weighted, ADC, and high b-value diffusion MRI streams separate until a gating stage and trains with random modality dropout. This setup produces higher ranking scores under seven different missing-modality and artifact conditions across UNet, nnUNet, and Mamba backbones, with gains ranging from 2.8 percent to 13.4 percent. The improvement holds because each stream processes its own data independently and the dropout forces the model to learn compensation rather than relying on any single channel. In clinical terms this matters because diffusion sequences are the ones most often corrupted by motion or artifacts, so the method supports more reliable cancer detection when input quality varies. External tests further show that removing the ADC map improves performance on data from another institution, pointing to limited portability of computed maps.

Core claim

MIGF maintains separate modality-specific encoding streams before a learned gating stage and pairs this with modality dropout training. The resulting models improve ideal-scenario ranking scores by 2.8 percent for UNet, 4.6 percent for nnUNet, and 13.4 percent for Mamba; the best configuration reaches 0.7304 plus or minus 0.056. Robustness arises from strict isolation plus dropout-driven compensation, not from adaptive per-sample quality routing, because the gate converges to a stable modality prior. Missing T2-weighted images remain a shared failure mode, while tolerance to HBV and ADC degradation rises. On external Prostate158 data, domain shift is driven mainly by ADC map incompatibility,

What carries the argument

The Modality-Isolated Gated Fusion (MIGF) module, which runs independent modality encoding streams before a learned gate and trains with modality dropout to enforce compensation under incomplete inputs.

If this is right

  • MIGF improves tolerance specifically to degradation in HBV and ADC sequences while missing T2-weighted images stay a common failure point.
  • The gating stage converges to a stable modality weighting rather than performing sample-by-sample quality routing.
  • Deep supervision improves results only when used with the largest backbone models.
  • Excluding the ADC map raises performance on external data, showing that computed diffusion maps transfer less well than directly acquired sequences.

Where Pith is reading between the lines

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

  • The same isolation-plus-dropout pattern could be tested on other multi-modal medical imaging problems where one modality is frequently unreliable.
  • The stable prior learned by the gate suggests the method may work even without per-sample adaptation, which could simplify deployment.
  • Future experiments that vary artifact types beyond the seven tested scenarios would check whether the compensation effect generalizes.

Load-bearing premise

The observed robustness gains are produced by modality isolation and dropout rather than by backbone-specific tuning choices or the particular seven missing-modality scenarios that were tested.

What would settle it

Ablating either the modality isolation or the dropout training while keeping all other elements fixed and measuring whether the robustness gains under missing HBV or ADC conditions disappear would settle the causal claim.

Figures

Figures reproduced from arXiv: 2604.10702 by Aijing Luo, Kewen Chen, Luo Lei, Shanhu Yao, Wenzhao Xie, Yongbo Shu, Zirui Xin.

Figure 1
Figure 1. Figure 1: Overview of the proposed Modality-Isolated Gated Fusion (MIGF) framework. [PITH_FULL_IMAGE:figures/full_fig_p012_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Detailed structure of the MIGF module. Given modality-specific feature maps [PITH_FULL_IMAGE:figures/full_fig_p014_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Robustness profiles of bare and MIGF-equipped backbones across seven evalu [PITH_FULL_IMAGE:figures/full_fig_p023_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Demystifying Modality Robustness — The Interplay of Feature Isolation and Modality Dropout [PITH_FULL_IMAGE:figures/full_fig_p034_4.png] view at source ↗
read the original abstract

Multi-parametric prostate MRI combines T2-weighted (T2W), apparent diffusion coefficient (ADC), and high b-value diffusion-weighted (HBV) sequences for non-invasive detection of clinically significant prostate cancer. In practice, the diffusion sequences are more frequently subject to acquisition variability, motion, and artifacts than T2W, making robust fusion of these channels the clinically relevant requirement. We propose Modality-Isolated Gated Fusion (MIGF), an architecture-agnostic module that maintains separate modality-specific encoding streams before a learned gating stage, combined with modality dropout training to enforce compensation under incomplete inputs. We benchmark six backbones and assess MIGF-equipped models under seven missing-modality and artifact scenarios on PI-CAI (1,500 studies, fold-0 split, five seeds). MIGF improved ideal-scenario Ranking Score for UNet, nnUNet, and Mamba by 2.8%, 4.6%, and 13.4%; the best model (MIGFNet-nnUNet, gating + ModDrop, no deep supervision) achieved 0.7304 +/- 0.056. MIGF primarily improved tolerance to HBV/ADC degradation, while missing T2W remained a shared failure mode across all architectures. Mechanistic analysis shows that robustness gains arise from strict modality isolation and dropout-driven compensation rather than adaptive per-sample quality routing: the gate converged to a stable modality prior, and deep supervision was beneficial only for the largest backbone. External evaluation on Prostate158 (n=158) revealed substantial domain shift driven primarily by ADC map incompatibility across institutions; modality-isolation analysis confirmed that removing ADC improved external performance, identifying computed diffusion maps as inherently less portable than raw-acquired sequences.

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

3 major / 2 minor

Summary. The paper proposes Modality-Isolated Gated Fusion (MIGF), an architecture-agnostic module for multi-parametric prostate MRI segmentation that keeps separate modality-specific encoding streams before a learned gating stage and uses modality dropout during training to promote compensation under incomplete inputs. It benchmarks six backbones on the PI-CAI dataset (1,500 studies, fold-0 split, five seeds) across seven missing-modality and artifact scenarios, reporting ranking-score gains of 2.8%, 4.6%, and 13.4% for UNet, nnUNet, and Mamba in the ideal case; the top model (MIGFNet-nnUNet) reaches 0.7304 ± 0.056. External evaluation on Prostate158 (n=158) identifies domain shift driven by ADC incompatibility, with modality-isolation analysis showing that dropping ADC improves external performance. The central mechanistic claim is that gains stem from strict isolation plus dropout-driven compensation rather than adaptive per-sample routing, because the gate converges to a stable modality prior.

Significance. If the attribution to modality isolation and dropout is substantiated, the work supplies a practical, backbone-agnostic plug-in that improves robustness to the clinically common problem of degraded diffusion sequences in prostate MRI without requiring architecture redesign. The evaluation design—six backbones, seven scenarios, five seeds, and an external cohort—provides concrete, reproducible numbers that go beyond single-backbone studies. Explicit credit is due for the external-cohort analysis that isolates ADC portability as a limiting factor and for the observation that T2W remains a shared failure mode across architectures.

major comments (3)
  1. [Abstract and §4] Abstract and §4 (mechanistic analysis): the claim that 'robustness gains arise from strict modality isolation and dropout-driven compensation rather than adaptive per-sample quality routing' and that 'the gate converged to a stable modality prior' is load-bearing for the central explanation, yet the manuscript supplies neither gate-output statistics (means, variances, or per-sample histograms across folds/seeds) nor an ablation that removes modality isolation while retaining dropout and the gating network.
  2. [§4.2 and Table 2] §4.2 and Table 2: the reported ranking-score improvements (2.8%, 4.6%, 13.4%) and the best-model score 0.7304 ± 0.056 are presented without p-values or confidence intervals on the differences versus the non-MIGF baselines, so it is impossible to judge whether the observed gains exceed what would be expected from random variation across the five seeds.
  3. [§4.3] §4.3 (external evaluation): the domain-shift analysis concludes that 'removing ADC improved external performance,' but the quantitative magnitude of the shift (e.g., Dice or ranking-score delta with vs. without ADC) is not reported, leaving the claim that computed diffusion maps are 'inherently less portable' unsupported by a direct numerical comparison.
minor comments (2)
  1. [Abstract] The abstract states 'MIGF primarily improved tolerance to HBV/ADC degradation' but does not define the exact ranking-score metric or the seven scenarios in sufficient detail for a reader to reproduce the experimental protocol without consulting the full methods section.
  2. [Figures] Figure captions and legends should explicitly state the number of seeds and the exact missing-modality combinations used for each bar or curve.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. These points help strengthen the mechanistic interpretation, statistical rigor, and external validation in the manuscript. We respond to each major comment below and will revise the paper accordingly.

read point-by-point responses

  1. Referee: [Abstract and §4] Abstract and §4 (mechanistic analysis): the claim that 'robustness gains arise from strict modality isolation and dropout-driven compensation rather than adaptive per-sample quality routing' and that 'the gate converged to a stable modality prior' is load-bearing for the central explanation, yet the manuscript supplies neither gate-output statistics (means, variances, or per-sample histograms across folds/seeds) nor an ablation that removes modality isolation while retaining dropout and the gating network.

    Authors: We agree that quantitative support for the gate behavior would strengthen the central claim. In the revised manuscript we will add gate-output statistics (means, standard deviations, and per-sample histograms) aggregated across the five seeds and folds. We will also conduct and report the requested ablation that retains dropout and the gating network but removes strict modality isolation (i.e., allowing early cross-modal mixing). This will directly test whether isolation is necessary for the observed robustness gains. revision: yes

  2. Referee: [§4.2 and Table 2] §4.2 and Table 2: the reported ranking-score improvements (2.8%, 4.6%, 13.4%) and the best-model score 0.7304 ± 0.056 are presented without p-values or confidence intervals on the differences versus the non-MIGF baselines, so it is impossible to judge whether the observed gains exceed what would be expected from random variation across the five seeds.

    Authors: We concur that formal statistical testing is required to assess whether the reported gains are distinguishable from seed-to-seed variability. In the revision we will compute and report paired statistical tests (e.g., Wilcoxon signed-rank or paired t-tests) together with 95% confidence intervals on the ranking-score differences between each MIGF variant and its corresponding baseline, using the five independent seeds as the unit of replication. revision: yes

  3. Referee: [§4.3] §4.3 (external evaluation): the domain-shift analysis concludes that 'removing ADC improved external performance,' but the quantitative magnitude of the shift (e.g., Dice or ranking-score delta with vs. without ADC) is not reported, leaving the claim that computed diffusion maps are 'inherently less portable' unsupported by a direct numerical comparison.

    Authors: We will revise §4.3 to include the exact numerical deltas. Specifically, we will report the change in Dice score and ranking score on Prostate158 when ADC is included versus excluded (while keeping T2W and HBV), thereby providing a direct quantitative comparison that supports the claim of reduced portability for computed diffusion maps. revision: yes

Circularity Check

0 steps flagged

No circularity: purely empirical method with experimental validation only

full rationale

The paper introduces MIGF as an architecture-agnostic module with modality-isolated streams, learned gating, and modality dropout training, then evaluates it via benchmarks on PI-CAI (fold-0, five seeds) and Prostate158 under seven missing-modality scenarios. All reported gains (e.g., 2.8–13.4% ranking score improvements) and mechanistic conclusions (gate convergence to stable prior, isolation plus dropout as operative factors) rest on direct experimental comparisons and ablation-style observations rather than any equations, derivations, or first-principles results. No self-definitional reductions, fitted inputs renamed as predictions, load-bearing self-citations, uniqueness theorems, or smuggled ansatzes appear; the central claims are falsifiable via the reported metrics and external evaluation. The derivation chain is therefore self-contained as standard empirical validation.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 1 invented entities

The central claim rests on empirical benchmarking rather than theoretical derivation. The method introduces one new module whose behavior is learned from data.

free parameters (2)
  • gating network weights
    Learned parameters that control per-modality contribution; their values are fitted during training on PI-CAI.
  • modality dropout probability
    Hyperparameter controlling how often entire modalities are dropped during training; value not stated in abstract.
axioms (2)
  • domain assumption Random modality dropout during training forces the model to learn compensation strategies that generalize to test-time missing inputs.
    Invoked to explain why dropout training improves robustness; no formal proof supplied.
  • domain assumption Separate modality-specific encoders preserve information that early fusion would lose under artifacted inputs.
    Core design premise of MIGF; tested indirectly via performance deltas.
invented entities (1)
  • MIGF module no independent evidence
    purpose: Architecture-agnostic fusion block that isolates modalities before gated combination.
    Newly proposed component whose independent validation outside this work is not provided.

pith-pipeline@v0.9.0 · 5639 in / 1624 out tokens · 39874 ms · 2026-05-10T15:54:13.803368+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

41 extracted references · 35 canonical work pages · 2 internal anchors

  1. [1]

    , author Makowski, M.R

    author Adams, L.C. , author Makowski, M.R. , author Engel, G. , author Rattunde, M. , author Busch, F. , author Asbach, P. , author Hamm, B. , author Bressem, K.K. , year 2022 . title Prostate158 -- an expert-annotated 3t mri dataset and algorithm for prostate cancer detection . journal Comput. Biol. Med. volume 148 , pages 105817 . :10.1016/j.compbiomed....

    work page doi:10.1016/j.compbiomed.2022.105817 2022

  2. [2]

    , author El-Shater Bosaily, A

    author Ahmed, H.U. , author El-Shater Bosaily, A. , author Brown, L.C. , author Gabe, R. , author Kaplan, R. , author Parmar, M.K. , author Collaco-Moraes, Y. , author Ward, K. , author Hindley, R.G. , author Freeman, A. , author Kirkham, A.P. , author Oldroyd, R. , author Parker, C. , author Emberton, M. , year 2017 . title Diagnostic accuracy of multi-p...

    work page doi:10.1016/s0140-6736(16)32401-1 2017

  3. [3]

    , author Wysock, J.S

    author Bjurlin, M.A. , author Wysock, J.S. , author Taneja, S.S. , year 2014 . title Optimization of prostate biopsy: Review of technique and complications . journal Urol. Clin. North Am. volume 41 , pages 299--313 . :10.1016/j.ucl.2014.01.011

    work page doi:10.1016/j.ucl.2014.01.011 2014

  4. [4]

    , author Adshead, J

    author Brizmohun Appayya, M. , author Adshead, J. , author Ahmed, H.U. , author Allen, C. , author Bainbridge, A. , author Barrett, T. , author Giganti, F. , author Graham, J. , author Haslam, P. , author Johnston, E.W. , author Kastner, C. , author Kirkham, A.P.S. , author Lipton, A. , author McNeill, A. , author Moniz, L. , author Moore, C.M. , author N...

    work page doi:10.1111/bju.14361 2018

  5. [5]

    MONAI: An open-source framework for deep learning in healthcare

    author Cardoso, M.J. , author Li, W. , author Brown, R. , author Ma, N. , author Kerfoot, E. , author Wang, Y. , author Murrey, B. , author Myronenko, A. , author Zhao, C. , author Yang, D. , author Nath, V. , author He, Y. , author Xu, Z. , author Hatamizadeh, A. , author Myronenko, A. , author Zhu, W. , author Liu, Y. , author Zheng, M. , author Tang, Y...

    work page internal anchor Pith review arXiv 2022

  6. [6]

    , author Dou, Q

    author Chen, C. , author Dou, Q. , author Jin, Y. , author Chen, H. , author Qin, J. , author Heng, P.A. , year 2019 . title Robust multimodal brain tumor segmentation via feature disentanglement and gated fusion , in: booktitle Medical Image Computing and Computer Assisted Intervention -- MICCAI 2019 , pp. pages 447--456 . :10.1007/978-3-030-32248-9_50

    work page doi:10.1007/978-3-030-32248-9_50 2019

  7. [7]

    , author Abdulkadir, A

    author Cicek, O. , author Abdulkadir, A. , author Lienkamp, S.S. , author Brox, T. , author Ronneberger, O. , year 2016 . title 3d u-net: Learning dense volumetric segmentation from sparse annotation , in: booktitle Medical Image Computing and Computer-Assisted Intervention -- MICCAI 2016 , pp. pages 424--432 . :10.1007/978-3-319-46723-8_49

    work page doi:10.1007/978-3-319-46723-8_49 2016

  8. [8]

    title Evaluation utilities for 3d detection and diagnosis in medical imaging [www document]

    author DIAGNijmegen , year 2022 . title Evaluation utilities for 3d detection and diagnosis in medical imaging [www document] . howpublished https://github.com/DIAGNijmegen/picai_eval . note Accessed: 2026-03-30

    2022

  9. [9]

    , author Yu, X

    author Ding, Y. , author Yu, X. , author Yang, Y. , year 2021 . title Rfnet: Region-aware fusion network for incomplete multi-modal brain tumor segmentation , in: booktitle Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , pp. pages 3975--3984

    2021

  10. [10]

    , author Gopinath, K

    author Dolz, J. , author Gopinath, K. , author Yuan, J. , author Lombaert, H. , author Desrosiers, C. , author Ben Ayed, I. , year 2019 . title Hyperdense-net: A hyper-densely connected cnn for multi-modal image segmentation . journal IEEE Trans. Med. Imaging volume 38 , pages 1116--1126 . :10.1109/TMI.2018.2878669

    work page doi:10.1109/tmi.2018.2878669 2019

  11. [11]

    , author Joutard, S

    author Dorent, R. , author Joutard, S. , author Modat, M. , author Ourselin, S. , author Vercauteren, T. , author Cardoso, M.J. , year 2019 . title A hetero-modal variational encoder-decoder for joint modality completion and segmentation , in: booktitle Medical Image Computing and Computer-Assisted Intervention -- MICCAI 2019 , pp. pages 74--82 . :10.1007...

    work page doi:10.1007/978-3-030-32245-8_9 2019

  12. [12]

    , author Kasivisvanathan, V

    author Giganti, F. , author Kasivisvanathan, V. , author Kirkham, A. , author Punwani, S. , author Emberton, M. , author Moore, C.M. , author Allen, C. , year 2022 . title Prostate mri quality: A critical review of the last 5 years and the role of the pi-qual score . journal Br. J. Radiol. volume 95 , pages 20210415 . :10.1259/bjr.20210415

    work page doi:10.1259/bjr.20210415 2022

  13. [13]

    , author Dao, T

    author Gu, A. , author Dao, T. , year 2024 . title Mamba: Linear-time sequence modeling with selective state spaces , in: booktitle First Conference on Language Modeling (COLM) . https://openreview.net/forum?id=tEYskw1VY2

    2024

  14. [14]

    Roth, and Daguang Xu

    author Hatamizadeh, A. , author Nath, V. , author Tang, Y. , author Yang, D. , author Roth, H.R. , author Xu, D. , year 2022 . title Swin unetr: Swin transformers for semantic segmentation of brain tumors in mri images , in: editor Crimi, A. , editor Bakas, S. (Eds.), booktitle Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries ,...

    work page doi:10.1007/978-3-031-08999-2_22 2022

  15. [15]

    , author Vargas, H.A

    author Hotker, A.M. , author Vargas, H.A. , author Donati, O.F. , year 2022 . title Abbreviated mr protocols in prostate mri . journal Life (Basel) volume 12 , pages 552 . :10.3390/life12040552

    work page doi:10.3390/life12040552 2022

  16. [16]

    , author Maillard, M

    author Hu, M. , author Maillard, M. , author Zhang, Y. , author Ciceri, T. , author La Barbera, G. , author Bloch, I. , author Gori, P. , year 2020 . title Knowledge distillation from multi-modal to mono-modal segmentation networks , in: editor Martel, A.L. , editor Abolmaesumi, P. , editor Stoyanov, D. , editor Mateus, D. , editor Zuluaga, M.A. , editor ...

    work page doi:10.1007/978-3-030-59710-8_75 2020

  17. [17]

    , author Ruan, S

    author Huang, L. , author Ruan, S. , author Decazes, P. , author Den ux, T. , year 2025 a. title Deep evidential fusion with uncertainty quantification and reliability learning for multimodal medical image segmentation . journal Inf. Fusion volume 113 , pages 102648 . https://www.sciencedirect.com/science/article/pii/S1566253524004263, :10.1016/j.inffus.2...

    work page doi:10.1016/j.inffus.2024.102648 2025

  18. [18]

    , author Ye, J

    author Huang, Z. , author Ye, J. , author Wang, H. , author Deng, Z. , et al., year 2025 b. title Revisiting model scaling with a u-net benchmark for 3d medical image segmentation . journal Sci. Rep. volume 15 , pages 15617 . :10.1038/s41598-025-15617-1

    work page doi:10.1038/s41598-025-15617-1 2025

  19. [19]

    Nature Methods 18(2), 203–211 (2021)

    author Isensee, F. , author Jaeger, P.F. , author Kohl, S.A.A. , author Petersen, J. , author Maier-Hein, K.H. , year 2021 . title nnu-net: A self-configuring method for deep learning-based biomedical image segmentation . journal Nat. Methods volume 18 , pages 203--211 . :10.1038/s41592-020-01008-z

    work page doi:10.1038/s41592-020-01008-z 2021

  20. [20]

    , author Wald, T

    author Isensee, F. , author Wald, T. , author Ulrich, C. , author Baumgartner, M. , author Roy, S. , author Maier-Hein, K. , author J \"a ger, P.F. , year 2024 . title nnu-net revisited: A call for rigorous validation in 3d medical image segmentation , in: booktitle Proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024 ,...

    2024

  21. [21]

    , author Rannikko, A.S

    author Kasivisvanathan, V. , author Rannikko, A.S. , author Borghi, M. , author Panebianco, V. , author Mynderse, L.A. , author Vaarala, M.H. , author Briganti, A. , author Budaus, L. , author Hellawell, G. , author Hindley, R.G. , author Roobol, M.J. , author Eggener, S. , author Ghei, M. , author Villers, A. , author Bladou, F. , author Jichlinski, P. ,...

    work page doi:10.1056/nejmoa1801993 2018

  22. [22]

    , author Baj, G

    author Kazaj, P.M. , author Baj, G. , author Salimi, Y. , author Stark, A.W. , author Valenzuela, W. , author Siontis, G.C.M. , author Zaidi, H. , author Reyes, M. , author Graeni, C. , author Shiri, I. , year 2025 . title From claims to evidence: A unified framework and critical analysis of cnn vs. transformer vs. mamba in medical image segmentation . jo...

    work page doi:10.48550/arxiv.2503.01306 2025

  23. [23]

    , author Xie, S

    author Lee, C.Y. , author Xie, S. , author Gallagher, P. , author Zhang, Z. , author Tu, Z. , year 2015 . title Deeply-supervised nets , in: editor Lebanon, G. , editor Vishwanathan, S.V.N. (Eds.), booktitle Proceedings of the Eighteenth International Conference on Artificial Intelligence and Statistics , publisher PMLR , address San Diego, California, US...

    2015

  24. [24]

    Weakly - supervised diagnosis and detection of breast cancer using deep multiple instance learning,

    author Li, C. , author Osman, Y.B.M. , author Huang, W. , author Xue, Z. , author Han, H. , author Zheng, H. , author Wang, S. , year 2023 . title Uncertainty-aware multi-parametric magnetic resonance image information fusion for 3d object segmentation , in: booktitle 2023 IEEE 20th International Symposium on Biomedical Imaging (ISBI) , pp. pages 1--4 . :...

    work page doi:10.1109/isbi53787.2023.10230478 2023

  25. [25]

    Focal Loss for Dense Object Detection , booktitle =

    author Lin, T.Y. , author Goyal, P. , author Girshick, R. , author He, K. , author Dollar, P. , year 2017 . title Focal loss for dense object detection , in: booktitle Proceedings of the IEEE International Conference on Computer Vision , pp. pages 2980--2988 . :10.1109/ICCV.2017.324

    work page doi:10.1109/iccv.2017.324 2017

  26. [26]

    , author Vellekoop, A

    author Loeb, S. , author Vellekoop, A. , author Ahmed, H.U. , author Catto, J. , author Emberton, M. , author Nam, R. , author Rosario, D.J. , author Scattoni, V. , author Lotan, Y. , year 2013 . title Systematic review of complications of prostate biopsy . journal European Urology volume 64 , pages 876--892 . :10.1016/j.eururo.2013.05.049

    work page doi:10.1016/j.eururo.2013.05.049 2013

  27. [27]

    Decoupled Weight Decay Regularization

    author Loshchilov, I. , author Hutter, F. , year 2019 . title Decoupled weight decay regularization . https://arxiv.org/abs/1711.05101, http://arxiv.org/abs/1711.05101 arXiv:1711.05101

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  28. [28]

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

    author Milletari, F. , author Navab, N. , author Ahmadi, S.A. , year 2016 . title V-net: Fully convolutional neural networks for volumetric medical image segmentation . https://arxiv.org/abs/1606.04797, http://arxiv.org/abs/1606.04797 arXiv:1606.04797

    work page Pith review arXiv 2016

  29. [29]

    , author Wolf, C

    author Neverova, N. , author Wolf, C. , author Taylor, G. , author Nebout, F. , year 2016 . title Moddrop: Adaptive multi-modal gesture recognition . journal IEEE Trans. Pattern Anal. Mach. Intell. volume 38 , pages 1692--1706 . https://doi.org/10.1109/TPAMI.2015.2461544, :10.1109/TPAMI.2015.2461544

    work page doi:10.1109/tpami.2015.2461544 2016

  30. [30]

    , author Radosa, C.G

    author Plodeck, V. , author Radosa, C.G. , author Hubner, H.M. , author Baldus, C. , author Borkowetz, A. , author Thomas, C. , author Kuhn, J.P. , author Laniado, M. , author Hoffmann, R.T. , author Platzek, I. , year 2020 . title Rectal gas-induced susceptibility artefacts on prostate diffusion-weighted mri with epi read-out at 3.0 t: Does a preparatory...

    work page doi:10.1007/s00261-020-02600-9 2020

  31. [31]

    , author Agrotis, G

    author Pooch, E.H.P. , author Agrotis, G. , author Cai, L. , author Emberton, M. , author Shah, T.T. , author Ahmed, H.U. , author Beets-Tan, R.G.H. , author Benson, S. , author Janssen, T. , author Schoots, I.G. , year 2026 . title Semi-supervised learning in prostate mri tumor detection approaches fully supervised performance on external validation . jo...

    work page doi:10.1007/s00330-026-12324-x 2026

  32. [32]

    In: Medical Image Compu ting and Computer-Assisted Intervention – MICCAI 2015

    author Ronneberger, O. , author Fischer, P. , author Brox, T. , year 2015 . title U-net: Convolutional networks for biomedical image segmentation , in: booktitle Medical Image Computing and Computer-Assisted Intervention -- MICCAI 2015 , pp. pages 234--241 . :10.1007/978-3-319-24574-4_28

    work page doi:10.1007/978-3-319-24574-4_28 2015

  33. [33]

    , author Bosma, J.S

    author Saha, A. , author Bosma, J.S. , author Twilt, J.J. , author van Ginneken, B. , author Bjartell, A. , author Padhani, A.R. , author Bonekamp, D. , author Villeirs, G. , author Salomon, G. , author Giannarini, G. , author Kalpathy-Cramer, J. , author Barentsz, J. , author Maier-Hein, K.H. , author Rusu, M. , author Rouviere, O. , author van den Bergh...

    work page doi:10.1016/s1470-2045(24)00220-1 2024

  34. [34]

    , author Sudre, C.H

    author Shaw, R. , author Sudre, C.H. , author Ourselin, S. , author Cardoso, M.J. , author Pemberton, H.G. , year 2021 . title A heteroscedastic uncertainty model for decoupling sources of mri image quality . journal Mach. Learn. Biomed. Imaging volume 1 , pages 1--23 . :10.59275/j.melba.2021-8678

    work page doi:10.59275/j.melba.2021-8678 2021

  35. [35]

    , author Ding, J

    author Sun, K. , author Ding, J. , author Li, Q. , author Chen, W. , author Zhang, H. , author Sun, J. , author Jiao, Z. , author Ni, X. , year 2024 . title Cmaf-net: A cross-modal attention fusion-based deep neural network for incomplete multi-modal brain tumor segmentation . journal Quant. Imaging Med. Surg. volume 14 , pages 4579--4604 . :10.21037/qims-24-9

    work page doi:10.21037/qims-24-9 2024

  36. [36]

    , author Rosenkrantz, A.B

    author Turkbey, B. , author Rosenkrantz, A.B. , author Haider, M.A. , author Padhani, A.R. , author Villeirs, G. , author Macura, K.J. , author Tempany, C.M. , author Choyke, P.L. , author Cornud, F. , author Margolis, D.J. , author Thoeny, H.C. , author Verma, S. , year 2019 . title Prostate imaging reporting and data system version 2.1: 2019 update of p...

    work page doi:10.1016/j.eururo.2019.02.033 2019

  37. [37]

    , author Chen, Y

    author Wang, H. , author Chen, Y. , author Ma, C. , author Avery, J. , author Hull, L. , author Carneiro, G. , year 2023 . title Multi-modal learning with missing modality via shared-specific feature modelling , in: booktitle Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pp. pages 15878--15887

    2023

  38. [38]

    , author Zhang, F

    author Wu, B. , author Zhang, F. , author Xu, L. , author Shen, S. , author Shao, P. , author Sun, M. , author Liu, P. , author Yao, P. , author Xu, R.X. , year 2023 . title Modality preserving u-net for segmentation of multimodal medical images . journal Quant. Imaging Med. Surg. volume 13 , pages 5242--5257 . :10.21037/qims-22-1367

    work page doi:10.21037/qims-22-1367 2023

  39. [39]

    , author Zhang, J

    author Xing, J. , author Zhang, J. , year 2025 . title Segmentation of brain tumors using a multi-modal segment anything model (msam) with missing modality adaptation . journal Bioengineering (Basel) volume 12 , pages 871 . :10.3390/bioengineering12080871

    work page doi:10.3390/bioengineering12080871 2025

  40. [40]

    , author He, N

    author Zhang, Y. , author He, N. , author Yang, J. , author Li, Y. , author Wei, D. , author Huang, Y. , author Zhang, Y. , author He, Z. , author Zheng, Y. , year 2022 . title mmformer: Multimodal medical transformer for incomplete multimodal learning of brain tumor segmentation , in: editor Wang, L. , editor Dou, Q. , editor Fletcher, P.T. , editor Spei...

    work page doi:10.1007/978-3-031-16443-9_11 2022

  41. [41]

    , author Li, S

    author Zhao, J. , author Li, S. , year 2024 . title Evidence modeling for reliability learning and interpretable decision-making under multi-modality medical image segmentation . journal Comput. Med. Imaging Graph. volume 116 , pages 102422 . :10.1016/j.compmedimag.2024.102422

    work page doi:10.1016/j.compmedimag.2024.102422 2024