Pith. sign in

REVIEW 4 major objections 4 minor 30 references

Segmentation Regularized Training for Multi-Domain Deep Learning Registration applied to MR-Guided Prostate Cancer Radiotherapy

T0 review · 4 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A deep learning registration model trained on 3T MR-Sim scans aligns prostate MR images from 1.5T MR-Linac and 3T scanners without retraining.

desk verdict A useful, honest extension of ProRSeg to prostate MRgART, undercut by a missing ablation of the very loss the title credits for generalization. read the letter →

arxiv 2507.06966 v1 pith:D3GQYQZA submitted 2025-07-09 cs.CV physics.med-ph

Sudharsan Madhavan , Chengcheng Gui , Lando Bosma , Josiah Simeth , Jue Jiang , Nicolas Cote , Nima Hassan Rezaeian , Himanshu Nagar
show 3 more authors
Victoria Brennan Neelam Tyagi Harini Veeraraghavan
This is my paper
classification cs.CVphysics.med-ph
keywords deformableimageregistrationMR-guidedadaptiveradiotherapyprostatecancermulti-domaingeneralizationsegmentationconsistencylossdoseaccumulationMR-Linac
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper trains a contour-guided deep learning registration model, ProRSeg, on 3 Tesla MR simulation scans of prostate cancer patients and then tests it, without retraining, on same-domain, cross-domain, and mixed-domain MR data. Its central claim is that a weighted segmentation consistency loss makes the learned registration anatomically plausible across scanner types: bladder Dice scores are statistically unchanged across same, cross, and mixed domains, while rectum and clinical target volume accuracy is highest on the 1.5T MR-Linac pairs. If correct, a single model could propagate contours and support deformable dose accumulation for MR-guided adaptive radiotherapy across different MR platforms, avoiding per-site retraining. The paper also reports preliminary dose accumulation for 42 five-fraction patients, finding that 83.3% met the CTV coverage constraint $D95 \geq 40.0$ Gy and the bladder sparing constraint $D50 \leq 20.0$ Gy, while only 9.5% stayed under the mean-dose upper limit.

What carries the argument

The central object is ProRSeg, a progressively refined joint registration and segmentation framework. A recurrent registration network computes deformation vector fields through eight convolutional long short-term memory iterations while a recurrent segmentation network runs in parallel; at each iteration a spatial transform network propagates contours, and a weighted segmentation consistency loss compares those propagated contours with target segmentations using Dice overlap, with organ weights of 0.40 for bladder, 0.30 for rectum, and 0.30 for prostate CTV. Because the loss is applied at every refinement step, it provides deep supervision that steers the registration toward anatomically plausible solutions. After training, only the registration network is retained for testing, and no contour guidance is needed at inference time.

What would settle it

Recompute same-domain Dice scores after splitting training and test sets strictly by patient; if any patient appears in both sets, the reported bladder DSC of 0.88 would not establish same-domain generalization. A second check is to compare ProRSeg's mixed-domain MRSim-to-MRL registrations against expert-edited contours in a multi-institutional cohort.

Watch

Extended reading notes

Core claim

On held-out data, ProRSeg registered MR-MR pairs with high overlap: for the bladder, mean Dice Similarity Coefficient was 0.88 for same-domain (3T MR-Sim to 3T MR-Sim), 0.87 for cross-domain (1.5T MR-Linac to 1.5T MR-Linac), and 0.86 for mixed-domain (MR-Sim to MR-Linac), with no statistically significant differences across domains. For the rectum and CTV, performance was domain-dependent: the cross-domain MR-Linac setting gave the highest mean DSCs (0.89 for both), the same-domain setting was intermediate, and the mixed-domain MRSim-to-MRL setting was the most challenging. On same-domain data, ProRSeg outperformed rigid registration and two intensity-based iterative methods, SyN and EVolution, especially for the bladder and rectum, which undergo large deformations. The authors then used ProRSeg's deformation fields for deformable dose accumulation in 42 patients undergoing five-fraction MR-guided adaptive radiotherapy, showing that most patients met institutional CTV coverage and bladder sparing constraints but that the accumulated mean target dose usually exceeded the upper limit, which they interpret as a sign that current planning protocols rarely escalate dose.

Load-bearing premise

The central claim rests on the assumption that the 58 same-domain test image pairs do not share patients with the 262 training pairs; the paper does not state that the split was done at the patient level.

Editorial extensions

If this is right

  • A single ProRSeg model trained once on 3T MR-Sim images can propagate prostate CTV, bladder, and rectum contours on daily 1.5T MR-Linac images without fine-tuning.
  • Bladder contour propagation is domain-invariant in this setting, while rectum and CTV accuracy depends on which domains are being bridged, with MRSim-to-MRL registration as the weakest link.
  • Contour-guided deep learning registration can beat intensity-based iterative registration for large-deformation organs such as the bladder and rectum.
  • Deformable dose accumulation from ProRSeg is feasible for checking whether delivered five-fraction treatments meet institutional dose constraints.
  • Accumulated-dose analysis can expose protocol-level patterns, such as adequate CTV coverage and bladder sparing in 83.3% of patients but rare compliance with the mean-dose upper limit.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Because the paper does not state whether the same-domain test split was done by patient, the reported same-domain Dice scores could be optimistic if any test pairs share patients with training pairs; a patient-level split would settle this.
  • The cross-domain advantage for rectum and CTV may reflect task difficulty rather than scanner invariance: daily MR-Linac pairs have less anatomical drift than longitudinal 3T MR-Sim follow-up pairs.
  • One could test whether increasing the rectum or CTV weight in the segmentation consistency loss narrows the mixed-domain gap, and whether the same weighting strategy transfers to other pelvic or abdominal sites with large-deformation organs.
  • One could check whether the low 9.5% upper-limit compliance is a planning protocol effect rather than a registration artifact by recomputing accumulated doses with an independent deformable registration method.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The paper proposes ProRSeg, a recurrent joint registration-segmentation deep learning method for deformable MR-to-MR registration in prostate cancer radiotherapy. The model is trained on 262 pairs of 3T MR simulation scans with a weighted segmentation consistency loss and is evaluated on same-domain (MRSim-MRSim), cross-domain (MRL-MRL), and mixed-domain (MRSim-MRL) test sets. Contour propagation accuracy is measured with DSC, HD95, and MDA, and ProRSeg is compared to rigid, SyN, and EVO on the same-domain test set. A dose accumulation feasibility study on 42 patients derives DVHs and reports compliance with institutional dose constraints.

Significance. If the domain-generalization claim holds, the work could enable contour propagation and dose accumulation across 1.5T MR-Linac and 3T MR-Sim images without retraining, which would be clinically valuable. The study uses held-out cross-domain patients, multiple geometric metrics, and established registration baselines, and the authors are explicit about several limitations. However, the central attribution of the results to the segmentation consistency loss is not isolated by ablation, the same-domain test split is not described at patient level, and the dose accumulation results are not validated against a reference standard. These gaps currently prevent the stronger claims in the abstract and conclusion from being fully supported.

major comments (4)
  1. [Section 2.1] The manuscript does not state whether the 58 same-domain test pairs come from patients excluded from the 262 training pairs. Because the discovery dataset consists of 320 image pairs from 34 patients, pairs from the same patient could appear in both sets. If so, the same-domain DSCs in Table 1 and the comparisons against SyN and EVO in Section 3.1 are optimistically biased by potential leakage. Please report the number of distinct patients in each split or perform a strict patient-level split, and re-evaluate the same-domain results accordingly.
  2. [Sections 2.2 and 3.3] The conclusion that the weighted segmentation consistency loss 'significantly enhances cross-domain generalization' is not supported by any experiment that varies this loss. ProRSeg is compared only to rigid, SyN, and EVO, which differ in architecture, optimization, and use of contour information, so the accuracy gap cannot be attributed to the segmentation regularization. Furthermore, the benchmark MRL-trained model in Section 3.3 achieves significantly higher bladder DSC (0.91 vs 0.87, p < 0.001), indicating a residual domain gap. An ablation with lambda_cons = 0 or with equal organ weights is needed to isolate the contribution of the segmentation consistency loss.
  3. [Table 1 and Discussion] The cross-domain MRL-MRL test set is a daily-fraction task with limited anatomical change, whereas the same-domain MRSim-MRSim test set uses longitudinal follow-up scans spanning up to 24 months. The statistically similar bladder DSCs across domains (0.88, 0.87, 0.86) may therefore reflect differences in task difficulty rather than domain invariance. The Discussion acknowledges this confounder, but the Abstract and Conclusion present the results as evidence of multi-domain generalization. Please control for anatomical variability (for example, by comparing against a same-domain daily-fraction subset) or temper the claim to 'comparable accuracy on a less variable daily-fraction task'.
  4. [Section 3.4] The dose accumulation and constraint-compliance results are not validated against a reference standard. There is no comparison to accumulation using manually verified contours, an alternative DIR algorithm, or a deformable phantom with known ground-truth displacements. The manuscript itself labels this analysis as a feasibility study, but the Conclusion states that the results 'support the feasibility' of dose accumulation. Please either add a validation experiment or restrict the conclusion to stating that dosimetric metrics can be computed with the current workflow, with accuracy remaining unvalidated.
minor comments (4)
  1. [Figure 3 caption] The caption appears to swap the mixed- and cross-domain labels relative to the definitions in Section 2.1: MRL-MRL is the cross-domain case and MRSim-MRL is the mixed-domain case, but the caption reverses them.
  2. [Section 2.2.1] Several displayed equations are missing from the manuscript text, including the image similarity loss and the total registration loss expression, which hinders reproducibility. Please ensure all equations are rendered.
  3. [Section 2.2.1 and Discussion] The organ-specific weights (bladder 0.40, rectum 0.30, prostate 0.30) appear only in the Discussion; they should be defined in the Methods section where the weighted segmentation consistency loss is introduced.
  4. [References] Reference [7] appears to be an incomplete institutional thesis entry; please provide full bibliographic details.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: held-out evaluation is independent of training and no fitted parameter is repackaged as a prediction.

full rationale

ProRSeg is trained on 262 MR-Sim pairs using a weighted DSC segmentation-consistency loss, with hyperparameters lambda_smooth=30 and lambda_cons=5 selected on a validation set; the reported DSCs are computed on held-out same-domain (58 pairs), cross-domain (72 MRL pairs), and mixed-domain (42 pairs) datasets. Although the training loss and test metric share the DSC form, the test contours are not used for training or hyperparameter selection, so this is standard supervised evaluation rather than circularity. The paper adopts ProRSeg from the same group's prior work [10], but the architecture and loss are specified in the present paper and the method is independently compared with rigid, SyN, and EVO on the held-out data; the self-citation is provenance, not load-bearing support. The Discussion's stated limitations acknowledge the preliminary nature of the dose-accumulation feasibility analysis, and the unsupported causal claim that segmentation regularization 'significantly enhances cross-domain generalization' (no lambda_cons=0 ablation) is an experimental-support gap, not a circular reduction. Likewise, the absence of an explicit patient-level split in Section 2.1 raises a possible data-leakage concern for the same-domain comparison, but contamination is not equivalence-by-construction. No equation defines its target in terms of the quantity it is used to predict, and no fitted parameter is renamed as a prediction.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central claims rest on five main premises: accurate clinical contours, valid DVF dose mapping, the model's capacity to learn domain-invariant features from 3T alone, the independence of the same-domain test split, and the statistical tests. Three free parameter groups (organ weights, loss weights, preprocessing percentiles) are tuned by hand or grid search on validation, not on test outcomes. No invented entities are introduced.

free parameters (3)
  • Segmentation consistency loss organ weights = bladder 0.40, rectum 0.30, prostate 0.30
    Chosen by hand in Discussion to emphasize bladder deformation; not fit to test outcomes, but tune the central training loss.
  • lambda_smooth and lambda_cons = 30 and 5
    Selected by grid search on validation set (Section 2.2.2); standard hyperparameter tuning, affects final model.
  • Intensity clipping percentiles = 10th and 90th percentiles
    Preprocessing choice (Section 2.1.1) for normalization; fixed across domains, could affect generalization.
assumptions (5)
  • domain assumption The clinical segmentation masks used as training labels and evaluation targets are anatomically accurate ground truth.
    All training and DSC evaluation assume the manual/reference contours are correct; errors in contours would propagate to registration training and evaluation (Sections 2.1, 2.3).
  • domain assumption DVF-based dose mapping with trilinear interpolation correctly approximates delivered dose accumulation across fractions.
    Section 2.3.2 maps each fraction's dose with the registration DVF; no dosimetric validation against a gold standard is presented, and the paper calls this a feasibility study.
  • ad hoc to paper A recurrent CLSTM network with intensity normalization can learn domain-invariant features from 3T MRSim training sufficient for 1.5T MRL images.
    This is the core modeling assumption behind the cross-domain generalization claim; the paper does not ablate other domain adaptation strategies.
  • ad hoc to paper Held-out same-domain test pairs are independent of training patients.
    Section 2.1 splits 320 pairs into 262 training and 58 test without stating whether the split is by patient; if the same patient appears in both, the held-out estimate is optimistic.
  • standard math The statistical tests (paired/unpaired Wilcoxon) are appropriate for the small, paired samples.
    Standard nonparametric tests used in Section 2.3.3.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Segmentation Regularized Training for Multi-Domain Deep Learning Registration applied to MR-Guided Prostate Cancer Radiotherapy." pith.science (2026). https://pith.science/paper/D3GQYQZA

@misc{pith2026250706966,
  author       = {Pith},
  title        = {Pith review of: Segmentation Regularized Training for Multi-Domain Deep Learning Registration applied to MR-Guided Prostate Cancer Radiotherapy},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/D3GQYQZA}},
  note         = {Machine review of arXiv:2507.06966}
}
read the original abstract

Background: Accurate deformable image registration (DIR) is required for contour propagation and dose accumulation in MR-guided adaptive radiotherapy (MRgART). This study trained and evaluated a deep learning DIR method for domain invariant MR-MR registration. Methods: A progressively refined registration and segmentation (ProRSeg) method was trained with 262 pairs of 3T MR simulation scans from prostate cancer patients using weighted segmentation consistency loss. ProRSeg was tested on same- (58 pairs), cross- (72 1.5T MR Linac pairs), and mixed-domain (42 MRSim-MRL pairs) datasets for contour propagation accuracy of clinical target volume (CTV), bladder, and rectum. Dose accumulation was performed for 42 patients undergoing 5-fraction MRgART. Results: ProRSeg demonstrated generalization for bladder with similar Dice Similarity Coefficients across domains (0.88, 0.87, 0.86). For rectum and CTV, performance was domain-dependent with higher accuracy on cross-domain MRL dataset (DSCs 0.89) versus same-domain data. The model's strong cross-domain performance prompted us to study the feasibility of using it for dose accumulation. Dose accumulation showed 83.3% of patients met CTV coverage (D95 >= 40.0 Gy) and bladder sparing (D50 <= 20.0 Gy) constraints. All patients achieved minimum mean target dose (>40.4 Gy), but only 9.5% remained under upper limit (<42.0 Gy). Conclusions: ProRSeg showed reasonable multi-domain MR-MR registration performance for prostate cancer patients with preliminary feasibility for evaluating treatment compliance to clinical constraints.

Figures

Figures reproduced from arXiv: 2507.06966 by the authors.

Figure 2
Figure 2. Quantitative comparison of contour alignment produced using the DIR methods for [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Representative cases showing results of ProRSeg together with Jacobian determinant [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Dose-volume histograms (DVHs) derived using dose accumulation performed with ProRSeg for bladder, CTV, rectum, and urethra. Mean (solid lines), ± one standard deviation (light shaded areas), and min-max range (lightest shaded regions) are shown. Dose-volume histogram analysis derived from ProRSeg-based deformable registration revealed critical insights into organ-specific dose distributions ( [PITH_FULL_IMAGE:figur… view at source ↗
Figures from the paper (1 more)
Figure 5
Figure 5. Figure 5: Compliance rates for institutional dose constraints based on the accumulated dose distributions (n = 42 patients) using ProRSeg for CTV and bladder. Assessment of institutional dose constraint compliance following deformable dose accumulation revealed important insight…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

30 extracted references · 22 canonical work pages

  1. [1]

    Clinical implementation of magnetic resonance imaging guided adaptive radiotherapy for localized prostate cancer

    Tetar SU, Bruynzeel AME, Lagerwaard FJ, Slotman BJ, Bohoudi O, Palacios MA. Clinical implementation of magnetic resonance imaging guided adaptive radiotherapy for localized prostate cancer. Physics and Imaging in Radiation Oncology 2019;9:69–76. https://doi.org/10.1016/j.phro.2019.02.002

    work page doi:10.1016/j.phro.2019.02.002 2019

  2. [2]

    Magnetic Resonance Imaging–Guided vs Computed Tomography–Guided Stereotactic Body Radiotherapy for Prostate Cancer: The MIRAGE Randomized Clinical Trial

    Kishan AU, Ma TM, Lamb JM, Casado M, Wilhalme H, Low DA, et al. Magnetic Resonance Imaging–Guided vs Computed Tomography–Guided Stereotactic Body Radiotherapy for Prostate Cancer: The MIRAGE Randomized Clinical Trial. JAMA Oncology 2023;9:365–73. https://doi.org/10.1001/jamaoncol.2022.6558

    work page arXiv 2023

  3. [3]

    Use of image registration and fusion algorithms and techniques in radiotherapy: Report of the AAPM Radiation Therapy Committee Task Group No

    Brock KK, Mutic S, McNutt TR, Li H, Kessler ML. Use of image registration and fusion algorithms and techniques in radiotherapy: Report of the AAPM Radiation Therapy Committee Task Group No. 132. Medical Physics 2017;44:e43–76. https://doi.org/10.1002/mp.12256

    work page doi:10.1002/mp.12256 2017

  4. [4]

    End-to-end empirical validation of dose accumulation in MRI-guided adaptive radiotherapy for prostate cancer using an anthropomorphic deformable pelvis phantom

    Bohoudi O, Lagerwaard FJ, Bruynzeel AME, Niebuhr NI, Johnen W, Senan S, et al. End-to-end empirical validation of dose accumulation in MRI-guided adaptive radiotherapy for prostate cancer using an anthropomorphic deformable pelvis phantom. Radiotherapy and Oncology 2019;141:200–

    work page 2019

  5. [5]

    Validation of an MR-guided online adaptive radiotherapy (MRgoART) program: Deformation accuracy in a heterogeneous, deformable, anthropomorphic phantom

    Mittauer KE, Hill PM, Bassetti MF, Bayouth JE. Validation of an MR-guided online adaptive radiotherapy (MRgoART) program: Deformation accuracy in a heterogeneous, deformable, anthropomorphic phantom. Radiotherapy and Oncology 2020;146:97–109. https://doi.org/10.1016/j.radonc.2020.02.012

    work page doi:10.1016/j.radonc.2020.02.012 2020

  6. [6]

    A multi-institutional comparison of retrospective deformable dose accumulation for online adaptive magnetic resonance- guided radiotherapy

    Murr M, Bernchou U, Bubula-Rehm E, Ruschin M, Sadeghi P, Voet P, et al. A multi-institutional comparison of retrospective deformable dose accumulation for online adaptive magnetic resonance- guided radiotherapy. Physics and Imaging in Radiation Oncology 2024;30:100588. https://doi.org/10.1016/j.phro.2024.100588

    work page arXiv 2024

  7. [7]

    https://doi.org/10.1016/j.radonc.2019.09.014

    work page doi:10.1016/j.radonc.2019.09.014 2019

  8. [8]

    Dose accumulation of adapted treatment plans in MR-guided radiotherapy

    Leiner L, Fakultät HU (Germany) M. Dose accumulation of adapted treatment plans in MR-guided radiotherapy. Dose accumulation of adapted treatment plans in MR-guided radiotherapy, 2024

    work page 2024

Show all 30 references
  1. [9]

    Geometric and Dosimetric Validation of Deformable Image Registration for Prostate MR-guided Adaptive Radiotherapy 2025

    Malkov VN, Mansour IR, Kong V, Li W, Dang J, Sadeghi P, et al. Geometric and Dosimetric Validation of Deformable Image Registration for Prostate MR-guided Adaptive Radiotherapy 2025. https://doi.org/10.48550/arXiv.2504.07933

    2025 doi

  2. [10]

    Zhang Y, Paulson E, Lim S, Hall WA, Ahunbay E, Mickevicius NJ, et al. A Patient-Specific Autosegmentation Strategy Using Multi-Input Deformable Image Registration for Magnetic Resonance Imaging–Guided Online Adaptive Radiation Therapy: A Feasibility Study. Advances in Radiatio...

    2020 doi

  3. [11]

    Progressively refined deep joint registration segmentation (ProRSeg) of gastrointestinal organs at risk: Application to MRI and cone-beam CT

    Jiang J, Hong J, Tringale K, Reyngold M, Crane C, Tyagi N, et al. Progressively refined deep joint registration segmentation (ProRSeg) of gastrointestinal organs at risk: Application to MRI and cone-beam CT. Medical Physics 2023;50:4758–74. https://doi.org/10.1002/mp.16527

    2023 doi

  4. [12]

    VoxelMorph: A Learning Framework for Deformable Medical Image Registration

    Balakrishnan G, Zhao A, Sabuncu MR, Guttag J, Dalca AV. VoxelMorph: A Learning Framework for Deformable Medical Image Registration. IEEE Trans Med Imaging 2019;38:1788–800. https://doi.org/10.1109/TMI.2019.2897538

    2019

  5. [13]

    A deep learning framework for unsupervised affine and deformable image registration

    de Vos BD, Berendsen FF, Viergever MA, Sokooti H, Staring M, Išgum I. A deep learning framework for unsupervised affine and deformable image registration. Medical Image Analysis 2019;52:128–43. https://doi.org/10.1016/j.media.2018.11.010

    2019 doi

  6. [14]

    Evaluation and mitigation of deformable image registration uncertainties for MRI-guided adaptive radiotherapy

    Zhong H, Kainz KK, Paulson ES. Evaluation and mitigation of deformable image registration uncertainties for MRI-guided adaptive radiotherapy. Journal of Applied Clinical Medical Physics 2024;25:e14358. https://doi.org/10.1002/acm2.14358

    2024 doi

  7. [15]

    Training Data Independent Image Registration with Gans Using Transfer Learning and Segmentation Information

    Mahapatra D, Ge Z. Training Data Independent Image Registration with Gans Using Transfer Learning and Segmentation Information. 2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019), 2019, p. 709–13. https://doi.org/10.1109/ISBI.2019.8759247

    2019

  8. [16]

    On the Adaptability of Unsupervised CNN-Based Deformable Image Registration to Unseen Image Domains

    Ferrante E, Oktay O, Glocker B, Milone DH. On the Adaptability of Unsupervised CNN-Based Deformable Image Registration to Unseen Image Domains. In: Shi Y, Suk H-I, Liu M, editors. Machine Learning in Medical Imaging, Cham: Springer International Publishing; 2018, p. 294–302. h...

    2018 doi

  9. [17]

    Integration of operator- validated contours in deformable image registration for dose accumulation in radiotherapy

    Bosma LS, Ries M, Denis de Senneville B, Raaymakers BW, Zachiu C. Integration of operator- validated contours in deformable image registration for dose accumulation in radiotherapy. Physics and Imaging in Radiation Oncology 2023;27:100483. https://doi.org/10.1016/j.phro.2023.100483

    2023

  10. [18]

    Anatomically-adaptive multi-modal image registration for image-guided external-beam radiotherapy

    Zachiu C, Denis de Senneville B, Willigenburg T, Voort van Zyp JRN, de Boer JCJ, Raaymakers BW, et al. Anatomically-adaptive multi-modal image registration for image-guided external-beam radiotherapy. Phys Med Biol 2020;65:215028. https://doi.org/10.1088/1361-6560/abad7d

    2020 doi

  11. [19]

    A contour-guided deformable image registration algorithm for adaptive radiotherapy

    Gu X, Dong B, Wang J, Yordy J, Mell L, Jia X, et al. A contour-guided deformable image registration algorithm for adaptive radiotherapy. Phys Med Biol 2013;58:1889. https://doi.org/10.1088/0031-9155/58/6/1889

    2013 doi

  12. [20]

    MuSIC: Multi- Sequential Interactive Co-Registration for Cancer Imaging Data based on Segmentation Masks

    Eichner T, Mörth E, Wagner-Larsen K, Lura N, Haldorsen I, Gröller E, et al. MuSIC: Multi- Sequential Interactive Co-Registration for Cancer Imaging Data based on Segmentation Masks. Eurographics Workshop on Visual Computing for Biology and Medicine 2022:81–91. https://doi.org/...

    2022 doi

  13. [21]

    Joint Registration and Segmentation via Multi-Task Learning for Adaptive Radiotherapy of Prostate Cancer

    Elmahdy MS, Beljaards L, Yousefi S, Sokooti H, Verbeek F, Van Der Heide UA, et al. Joint Registration and Segmentation via Multi-Task Learning for Adaptive Radiotherapy of Prostate Cancer. IEEE Access 2021;9:95551–68. https://doi.org/10.1109/ACCESS.2021.3091011

    2021

  14. [22]

    Semi-weakly-supervised neural network training for medical image registration 2024

    Li Y, Fu Y, Gayo IJMB, Yang Q, Min Z, Saeed SU, et al. Semi-weakly-supervised neural network training for medical image registration 2024. https://doi.org/10.48550/arXiv.2402.10728

    2024 doi arXiv

  15. [23]

    A Coupled Global Registration and Segmentation Framework With Application to Magnetic Resonance Prostate Imagery

    Gao Y, Sandhu R, Fichtinger G, Tannenbaum AR. A Coupled Global Registration and Segmentation Framework With Application to Magnetic Resonance Prostate Imagery. IEEE Transactions on Medical Imaging 2010;29:1781–94. https://doi.org/10.1109/TMI.2010.2052065

    2010

  16. [24]

    Contour-guided deep learning based deformable image registration for dose monitoring during CBCT-guided radiotherapy of prostate cancer

    Hemon C, Rigaud B, Barateau A, Tilquin F, Noblet V, Sarrut D, et al. Contour-guided deep learning based deformable image registration for dose monitoring during CBCT-guided radiotherapy of prostate cancer. Journal of Applied Clinical Medical Physics 2023;24:e13991. https://doi...

    2023 doi

  17. [25]

    SBRT focal dose intensification using an MR-Linac adaptive planning for intermediate-risk prostate cancer: An analysis of the dosimetric impact of intra-fractional organ changes

    Brennan VS, Burleson S, Kostrzewa C, Godoy Scripes P, Subashi E, Zhang Z, et al. SBRT focal dose intensification using an MR-Linac adaptive planning for intermediate-risk prostate cancer: An analysis of the dosimetric impact of intra-fractional organ changes. Radiotherapy and ...

    2023

  18. [26]

    The ANTsX ecosystem for quantitative biological and medical imaging

    Tustison NJ, Cook PA, Holbrook AJ, Johnson HJ, Muschelli J, Devenyi GA, et al. The ANTsX ecosystem for quantitative biological and medical imaging. Sci Rep 2021;11:9068. https://doi.org/10.1038/s41598-021-87564-6

    2021 doi

  19. [27]

    Symmetric diffeomorphic image registration with cross-correlation: Evaluating automated labeling of elderly and neurodegenerative brain

    Avants BB, Epstein CL, Grossman M, Gee JC. Symmetric diffeomorphic image registration with cross-correlation: Evaluating automated labeling of elderly and neurodegenerative brain. Medical Image Analysis 2008;12:26–41. https://doi.org/10.1016/j.media.2007.06.004

    2008 doi

  20. [28]

    EVolution: an edge-based variational method for non-rigid multi-modal image registration

    Denis de Senneville B, Zachiu C, Ries M, Moonen C. EVolution: an edge-based variational method for non-rigid multi-modal image registration. Phys Med Biol 2016;61:7377. https://doi.org/10.1088/0031-9155/61/20/7377

    2016 doi

  21. [29]

    Learning Deformable Image Registration with Structure Guidance Constraints for Adaptive Radiotherapy

    Kuckertz S, Papenberg N, Honegger J, Morgas T, Haas B, Heldmann S. Learning Deformable Image Registration with Structure Guidance Constraints for Adaptive Radiotherapy. In: Špiclin Ž, McClelland J, Kybic J, Goksel O, editors. Biomedical Image Registration, Cham: Springer Inter...

    2020 doi

  22. [30]

    Domain Adaptation for Medical Image Analysis: A Survey

    Guan H, Liu M. Domain Adaptation for Medical Image Analysis: A Survey. IEEE Transactions on Biomedical Engineering 2022;69:1173–85. https://doi.org/10.1109/TBME.2021.3117407

    2022

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.