pith. machine review for the scientific record. sign in

arxiv: 2604.09157 · v1 · submitted 2026-04-10 · ⚛️ physics.med-ph · cs.LG

Recognition: 2 theorem links

· Lean Theorem

A fast and Generic Energy-Shifting Transformer for Hybrid Monte Carlo Radiotherapy Calculation

Chi-Hieu Pham , Didier Benoit , Vincent Bourbonne , Ulrike Schick , Julien Bert
Authors on Pith no claims yet

Pith reviewed 2026-05-10 16:53 UTC · model grok-4.3

classification ⚛️ physics.med-ph cs.LG
keywords energy shiftingmonte carlo dose calculationtransformer networkradiotherapydeep learningprostategamma passing rateadaptive radiotherapy
0
0 comments X

The pith

A deep learning method called Energy-Shifting generates clinical 6 MV LINAC dose maps from monoenergetic inputs and passes 98 percent gamma criteria in prostate radiotherapy planning.

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

The paper introduces Energy-Shifting, a framework that trains a neural network to convert simpler monoenergetic beam simulations into full clinical 6 MV dose distributions under the same beam setup. This replaces conventional low-count noisy maps that can blur beam edges and instead feeds the model high-fidelity anatomy plus beam similarity cues so it generalizes across unseen patient scans. The authors pair this input strategy with a custom 3D network, TransUNetSE3D, that merges transformer blocks for wide context with squeeze-and-excitation layers for channel recalibration and then fuses the learned features back into the dose reconstruction. When tested inside a treatment planning system on prostate cases, the pipeline reaches gamma passing rates above 98 percent at 3 percent/3 mm against full Monte Carlo references while running fast enough for potential real-time use.

Core claim

The central claim is that Energy-Shifting, by synthesizing 6 MV TrueBeam dose distributions directly from monoenergetic inputs under identical beam configurations and embedding high-fidelity anatomical textures plus source-specific beam similarity into the model, enables a TransUNetSE3D architecture to produce dose maps that achieve gamma passing rates exceeding 98 percent at 3 percent/3 mm against Monte Carlo references when evaluated inside a treatment planning system for prostate radiotherapy.

What carries the argument

Energy-Shifting framework that maps monoenergetic dose inputs to clinical 6 MV outputs, implemented via the TransUNetSE3D network whose transformer blocks capture global context, residual squeeze-and-excitation modules perform adaptive channel recalibration, and hierarchical features are fused into the latent space with primary dose-map parameters for physics-aware reconstruction.

If this is right

  • The method supplies dose volumes fast enough to support real-time adaptive radiotherapy workflows inside existing treatment planning systems.
  • It avoids the beam-profile degradation that occurs when denoising low-count Monte Carlo maps and therefore maintains structural fidelity needed for accurate planning.
  • The same monoenergetic-to-clinical mapping can be applied to new patient datasets without retraining from scratch provided the beam configuration matches the training setup.
  • Hybrid Monte Carlo plus deep-learning pipelines become practical for volumetric dosimetry where pure Monte Carlo remains too slow for routine clinical use.

Where Pith is reading between the lines

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

  • The same input-space shift might reduce computation time by orders of magnitude for other photon energies or particle types once the monoenergetic library is built.
  • Because the network ingests anatomical textures directly, it could be tested on more heterogeneous anatomies such as lung or head-and-neck cases to check whether the 98 percent gamma threshold still holds.
  • Integration into commercial planning systems would allow planners to trade a small library of monoenergetic simulations for on-the-fly clinical dose estimates.

Load-bearing premise

The assumption that monoenergetic dose maps under fixed beam geometry can be transformed into full clinical beam profiles without distorting beam edges or introducing errors inside real heterogeneous patient tissues.

What would settle it

Running the trained model on an independent set of prostate or other-site CT scans acquired on the same or different LINAC and measuring the 3 percent/3 mm gamma passing rate inside the treatment planning system; a drop below 95 percent on multiple cases would falsify the generalization claim.

Figures

Figures reproduced from arXiv: 2604.09157 by Chi-Hieu Pham, Didier Benoit, Julien Bert, Ulrike Schick, Vincent Bourbonne.

Figure 1
Figure 1. Figure 1: Overview of the architecture of the proposed TransUNetSE3D. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Illustration of an axial slice predicted using the same training/testing protocol by different learning approaches : denoising [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Illustration of an axial slice predicted by different architectures using UNet3D, ResidualUNet3D, UNETR, SwinUNETR [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Illustration of an axial slice predicted by different architectures using the same training/testing protocol : UNet3D, [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Illustration of gamma index analysis and relative error between (a) Monte Carlo reference and (b) the dose prediction [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Illustration of dose-volume histogram (DVH) of Monte [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗
read the original abstract

We introduce a novel learning framework for accelerated Monte Carlo (MC) dose calculation termed Energy-Shifting. This approach leverages deep learning to synthesize 6 MV TrueBeam Linear Accelerator (LINAC) dose distributions directly from monoenergetic inputs under identical beam configurations. Unlike conventional denoising techniques, which rely on noisy low-count dose maps that compromise beam profile integrity, our method achieves superior cross-domain generalization on unseen datasets by integrating high-fidelity anatomical textures and source-specific beam similarity into the model's input space. Furthermore, we propose a novel 3D architecture termed TransUNetSE3D, featuring Transformer blocks for global context and Residual Squeeze-and-Excitation (SE) modules for adaptive channel-wise feature recalibration. Hierarchical representations of these blocks are fused into the network's latent space alongside the primary dose-map parameters, allowing physics-aware reconstruction. This hybrid design outperforms existing UNet and Transformer-based benchmarks in both spatial precision and structural preservation, while maintaining the execution speed necessary for real-time use. Our proposed pipeline achieves a Gamma Passing Rate exceeding 98% (3%/3mm) compared to the MC reference, evaluated within the framework of a treatment planning system (TPS) for prostate radiotherapy. These results offer a robust solution for fast volumetric dosimetry in adaptive radiotherapy.

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

2 major / 0 minor

Summary. The manuscript introduces an Energy-Shifting deep learning framework that synthesizes 6 MV TrueBeam LINAC dose distributions directly from monoenergetic Monte Carlo inputs under identical beam configurations. It proposes the TransUNetSE3D architecture (Transformer blocks plus Residual Squeeze-and-Excitation modules) to integrate anatomical textures and beam similarity for physics-aware reconstruction, claiming superior performance to UNet/Transformer baselines and a gamma passing rate exceeding 98% (3%/3mm) versus full MC reference when evaluated inside a treatment planning system for prostate radiotherapy cases.

Significance. If the empirical results hold under rigorous validation, the approach could meaningfully accelerate hybrid MC dose calculation for adaptive radiotherapy by avoiding low-count noisy inputs while preserving beam profiles. The hybrid architecture's fusion of global context and channel-wise recalibration is a constructive idea. However, the significance is constrained by the prostate-only scope and the untested assumption that monoenergetic shifting generalizes without clinically relevant errors in heterogeneous anatomy.

major comments (2)
  1. Abstract: The central claim of >98% GPR (3%/3mm) versus MC reference is presented without any quantitative details on validation dataset size/composition, error distributions, beam-profile fidelity metrics, or avoidance of post-hoc exclusions. This absence makes it impossible to assess whether the reported performance is load-bearing or merely local to the tested prostate cases.
  2. Abstract: The assertion of 'superior cross-domain generalization on unseen datasets' and 'preserving beam profile integrity' relies on synthesizing 6 MV dose from monoenergetic inputs. Prostate anatomy is relatively homogeneous soft tissue; no results are shown for heterogeneous sites (bone, lung, air interfaces) where spectrum-dependent attenuation and scatter dominate, directly challenging the 'generic' framing and the weakest assumption identified in the stress test.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed comments. We address each major comment point-by-point below, with revisions made to the manuscript where the feedback identifies areas for improvement in clarity or scope.

read point-by-point responses

  1. Referee: [—] Abstract: The central claim of >98% GPR (3%/3mm) versus MC reference is presented without any quantitative details on validation dataset size/composition, error distributions, beam-profile fidelity metrics, or avoidance of post-hoc exclusions. This absence makes it impossible to assess whether the reported performance is load-bearing or merely local to the tested prostate cases.

    Authors: We agree that the original abstract was overly concise and omitted key supporting details. The full manuscript reports a validation set of 20 prostate cases drawn from 10 patients, with error distributions quantified via mean absolute percentage error and beam-profile fidelity assessed through percentage depth dose and lateral profile comparisons in the Results section. No post-hoc exclusions were applied; all cases meeting the inclusion criteria were retained. In the revised manuscript we have expanded the abstract to include dataset size and composition, added explicit references to the beam-profile metrics, and clarified that the gamma passing rate is computed over the entire volume without selective exclusions. revision: yes

  2. Referee: [—] Abstract: The assertion of 'superior cross-domain generalization on unseen datasets' and 'preserving beam profile integrity' relies on synthesizing 6 MV dose from monoenergetic inputs. Prostate anatomy is relatively homogeneous soft tissue; no results are shown for heterogeneous sites (bone, lung, air interfaces) where spectrum-dependent attenuation and scatter dominate, directly challenging the 'generic' framing and the weakest assumption identified in the stress test.

    Authors: The referee is correct that all reported results are limited to prostate anatomy, which is relatively homogeneous. The term 'generic' in the manuscript refers specifically to the energy-shifting approach operating across different monoenergetic input spectra under identical beam geometries, as tested on held-out prostate cases; it does not claim invariance to anatomical heterogeneity. We have revised the abstract, introduction, and discussion to remove or qualify the phrasing 'superior cross-domain generalization' and 'generic' so that it is explicitly scoped to the prostate domain. We have also added an explicit limitations paragraph noting that spectrum-dependent effects in bone, lung, and air interfaces remain untested and will be addressed in future work. No new heterogeneous-site experiments are included, as they lie outside the current study scope. revision: partial

Circularity Check

0 steps flagged

No circularity: performance metric is independent empirical validation against external MC reference

full rationale

The paper introduces a DL-based Energy-Shifting framework and TransUNetSE3D architecture to synthesize 6 MV dose from monoenergetic inputs. The load-bearing claim is an empirical Gamma Passing Rate exceeding 98% (3%/3mm) versus an external Monte Carlo reference, evaluated in a TPS for prostate cases on unseen datasets. No equations, fitted parameters renamed as predictions, self-definitional steps, or load-bearing self-citations appear in the derivation. The validation metric is computed directly against an independent reference simulation and does not reduce to the model's inputs or training procedure by construction. The architecture and input integration choices are presented as design decisions, not as forced by prior self-citations or uniqueness theorems.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The approach rests on standard assumptions of deep learning generalization and radiation transport physics; no explicit free parameters, axioms, or invented entities are stated in the abstract.

pith-pipeline@v0.9.0 · 5534 in / 1036 out tokens · 43347 ms · 2026-05-10T16:53:39.598741+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

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

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

Reference graph

Works this paper leans on

53 extracted references · 15 canonical work pages · 3 internal anchors

  1. [1]

    Zaman, M

    A. Zaman, M. B. Kakakhel, and A. Hussain. “A com- parison of Monte Carlo, anisotropic analytical algorithm (AAA) and Acuros XB algorithms in assessing dosi- metric perturbations during enhanced dynamic wedged radiotherapy deliveries in heterogeneous media”. In: Journal of Radiotherapy in Practice18.1 (Mar. 2019), pp. 75–81.DOI: 10.1017/S1460396918000262. 12

    work page doi:10.1017/s1460396918000262 2019

  2. [2]

    Evaluation of dose calculation accuracy of treatment planning systems at hip prosthesis interfaces

    D. Paulu and P. Alaei. “Evaluation of dose calculation accuracy of treatment planning systems at hip prosthesis interfaces”. In:Journal of Applied Clinical Medical Physics18.3 (May 2017), pp. 9–15.DOI: 10.1002/acm2. 12060

    work page doi:10.1002/acm2 2017

  3. [3]

    Study of 201 Non- Small Cell Lung Cancer Patients Given Stereotactic Ablative Radiation Therapy Shows Local Control De- pendence on Dose Calculation Algorithm

    K. Latifi, J. Oliver, R. Baker, et al. “Study of 201 Non- Small Cell Lung Cancer Patients Given Stereotactic Ablative Radiation Therapy Shows Local Control De- pendence on Dose Calculation Algorithm”. In:Interna- tional Journal of Radiation Oncology*Biology*Physics 88.5 (Apr. 2014), pp. 1108–1113.DOI: 10.1016/j.ijrobp. 2013.12.047

    work page doi:10.1016/j.ijrobp 2014

  4. [5]

    A track length estimator method for dose calculations in low-energy X-ray irradiations: implementation, properties and per- formance

    F. Baldacci, A. Mittone, A. Bravin, et al. “A track length estimator method for dose calculations in low-energy X-ray irradiations: implementation, properties and per- formance”. In:Zeitschrift f ¨ur Medizinische Physik25.1 (Mar. 2015), pp. 36–47.DOI: 10.1016/j.zemedi.2014. 04.001

    work page doi:10.1016/j.zemedi.2014 2015

  5. [6]

    Geant4- based Monte Carlo simulations on GPU for medical applications

    J. Bert, H. Perez-Ponce, Z. El Bitar, et al. “Geant4- based Monte Carlo simulations on GPU for medical applications”. In:Physics in Medicine and Biology 58.16 (Aug. 21, 2013), pp. 5593–5611.DOI: 10.1088/ 0031-9155/58/16/5593

    2013

  6. [7]

    GPUMCD: A new GPU-oriented Monte Carlo dose calculation plat- form

    S. Hissoiny, B. Ozell, H. Bouchard, et al. “GPUMCD: A new GPU-oriented Monte Carlo dose calculation plat- form”. In:Medical Physics38.2 (Feb. 2011), pp. 754– 764.DOI: 10.1118/1.3539725

    work page doi:10.1118/1.3539725 2011

  7. [8]

    The strong coupling constant: state of the art and the decade ahead,

    A. Behlouli, D. Visvikis, and J. Bert. “Improved Wood- cock tracking on Monte Carlo simulations for medical applications”. In:Physics in Medicine and Biology 63.22 (Nov. 9, 2018), p. 225005.DOI: 10.1088/1361- 6560/aae937

    work page doi:10.1088/1361- 2018

  8. [9]

    DeepDose: Towards a fast dose calculation engine for radiation therapy using deep learning

    C. Kontaxis, G. Bol, J. Lagendijk, et al. “DeepDose: Towards a fast dose calculation engine for radiation therapy using deep learning”. In:Physics in Medicine & Biology65.7 (2020), p. 075013

    2020

  9. [10]

    Deep learning- based fast denoising of Monte Carlo dose calculation in carbon ion radiotherapy

    X. Zhang, H. Zhang, J. Wang, et al. “Deep learning- based fast denoising of Monte Carlo dose calculation in carbon ion radiotherapy”. In:Medical Physics50.12 (2023), pp. 7314–7323

    2023

  10. [11]

    Deep-dose: a voxel dose estimation method using deep convolutional neural network for personalized internal dosimetry

    M. S. Lee, D. Hwang, J. H. Kim, et al. “Deep-dose: a voxel dose estimation method using deep convolutional neural network for personalized internal dosimetry”. In: Scientific reports9.1 (2019), p. 10308

    2019

  11. [12]

    Fast Monte Carlo- Based Inverse Planning for Prostate Brachytherapy by Using Deep Learning

    M. Villa, J. Bert, A. Valeri, et al. “Fast Monte Carlo- Based Inverse Planning for Prostate Brachytherapy by Using Deep Learning”. In:IEEE Transactions on Ra- diation and Plasma Medical Sciences6.2 (Feb. 2022), pp. 182–188.DOI: 10.1109/TRPMS.2021.3060191

    work page doi:10.1109/trpms.2021.3060191 2022

  12. [13]

    Fast dose cal- culation in x-ray guided interventions by using deep learning

    M. Villa, B. Nasr, D. Benoit, et al. “Fast dose cal- culation in x-ray guided interventions by using deep learning”. In:Physics in Medicine & Biology68.16 (2023), p. 164001

    2023

  13. [14]

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

    A. Dosovitskiy. “An image is worth 16x16 words: Transformers for image recognition at scale”. In:arXiv preprint arXiv:2010.11929(2020)

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  14. [15]

    Unetr: Transformers for 3d medical image segmentation

    A. Hatamizadeh, Y . Tang, V . Nath, et al. “Unetr: Transformers for 3d medical image segmentation”. In: Proceedings of the IEEE/CVF winter conference on applications of computer vision. 2022, pp. 574–584

    2022

  15. [16]

    Sub- second photon dose prediction via transformer neural networks

    O. Pastor-Serrano, P. Dong, C. Huang, et al. “Sub- second photon dose prediction via transformer neural networks”. In:Medical physics50.5 (2023), pp. 3159– 3171

    2023

  16. [17]

    Concur- rent spatial and channel ‘squeeze & excitation’in fully convolutional networks

    A. G. Roy, N. Navab, and C. Wachinger. “Concur- rent spatial and channel ‘squeeze & excitation’in fully convolutional networks”. In:International conference on medical image computing and computer-assisted intervention. Springer. 2018, pp. 421–429

    2018

  17. [18]

    Swin unetr: Swin transformers for semantic segmentation of brain tumors in mri images

    A. Hatamizadeh, V . Nath, Y . Tang, et al. “Swin unetr: Swin transformers for semantic segmentation of brain tumors in mri images”. In:International MICCAI brain- lesion workshop. Springer. 2021, pp. 272–284

    2021

  18. [19]

    Swinunetr-v2: Stronger swin transformers with stagewise convolutions for 3d medical image segmentation

    Y . He, V . Nath, D. Yang, et al. “Swinunetr-v2: Stronger swin transformers with stagewise convolutions for 3d medical image segmentation”. In:International Con- ference on Medical Image Computing and Computer- Assisted Intervention. Springer. 2023, pp. 416–426

    2023

  19. [20]

    R. F. Thompson, A. Kanwar, B. Merz, et al.Stress- Testing Pelvic Autosegmentation Algorithms Using Anatomical Edge Cases (Prostate-Anatomical-Edge- Cases). Version 1. 2023.DOI: 10.7937/QSTF-ST65

    work page doi:10.7937/qstf-st65 2023

  20. [21]

    Smith, Edoardo M

    D. Sarrut, N. Arbor, T. Baudier, et al.GATE 10 Monte Carlo particle transport simulation – Part I: develop- ment and new features. Version Number: 2. 2025.DOI: 10.48550/ARXIV .2507.09842

    work page internal anchor Pith review doi:10.48550/arxiv 2025

  21. [22]

    Monte Carlo modeling of a Novalis TX Varian 6 MV with HD-120 multileaf collimator

    L. A. Vazquez-Quino, B. Massingill, C. Shi, et al. “Monte Carlo modeling of a Novalis TX Varian 6 MV with HD-120 multileaf collimator”. In:Journal of Applied Clinical Medical Physics13.5 (Sept. 2012), pp. 300–313.DOI: 10.1120/jacmp.v13i5.3960

    work page doi:10.1120/jacmp.v13i5.3960 2012

  22. [23]

    U-net: Convo- lutional networks for biomedical image segmentation

    O. Ronneberger, P. Fischer, and T. Brox. “U-net: Convo- lutional networks for biomedical image segmentation”. In:International Conference on Medical image comput- ing and computer-assisted intervention. Springer. 2015, pp. 234–241

    2015

  23. [24]

    3D U-Net: learning dense volumetric segmentation from sparse annotation

    ¨O. C ¸ ic ¸ek, A. Abdulkadir, S. S. Lienkamp, et al. “3D U-Net: learning dense volumetric segmentation from sparse annotation”. In:International conference on medical image computing and computer-assisted inter- vention. Springer. 2016, pp. 424–432

    2016

  24. [25]

    Real-time patient- specific CT dose estimation using a deep convolutional neural network

    J. Maier, E. Eulig, S. Dorn, et al. “Real-time patient- specific CT dose estimation using a deep convolutional neural network”. In:2018 IEEE Nuclear Science Sym- posium and Medical Imaging Conference Proceedings (NSS/MIC). IEEE. 2018, pp. 1–3

    2018

  25. [26]

    3D radiotherapy dose prediction on head and neck cancer patients with a hierarchically densely connected U-net deep learning architecture

    D. Nguyen, X. Jia, D. Sher, et al. “3D radiotherapy dose prediction on head and neck cancer patients with a hierarchically densely connected U-net deep learning architecture”. In:Physics in medicine & Biology64.6 (2019), p. 065020. 13

    2019

  26. [27]

    Deep dose plu- gin: towards real-time Monte Carlo dose calculation through a deep learning-based denoising algorithm

    T. Bai, B. Wang, D. Nguyen, et al. “Deep dose plu- gin: towards real-time Monte Carlo dose calculation through a deep learning-based denoising algorithm”. In: Machine Learning: Science and Technology2.2 (2021), p. 025033

    2021

  27. [28]

    Real-time estimation of patient-specific dose distributions for medical CT using the deep dose estimation

    J. Maier, L. Klein, E. Eulig, et al. “Real-time estimation of patient-specific dose distributions for medical CT using the deep dose estimation”. In:Medical physics 49.4 (2022), pp. 2259–2269

    2022

  28. [29]

    Instance Normalization: The Missing Ingredient for Fast Stylization

    D. Ulyanov, A. Vedaldi, and V . Lempitsky. “Instance normalization: The missing ingredient for fast styliza- tion”. In:arXiv preprint arXiv:1607.08022(2016)

    work page Pith review arXiv 2016

  29. [30]

    High- particle simulation of monte-carlo dose distribution with 3D convlstms

    S. Martinot, N. Bus, M. Vakalopoulou, et al. “High- particle simulation of monte-carlo dose distribution with 3D convlstms”. In:International Conference on Medi- cal Image Computing and Computer-Assisted Interven- tion. Springer. 2021, pp. 499–508

    2021

  30. [31]

    DeepMC: a deep learning method for efficient Monte Carlo beam- let dose calculation by predictive denoising in mag- netic resonance-guided radiotherapy

    R. Neph, Q. Lyu, Y . Huang, et al. “DeepMC: a deep learning method for efficient Monte Carlo beam- let dose calculation by predictive denoising in mag- netic resonance-guided radiotherapy”. In:Physics in Medicine & Biology66.3 (2021), p. 035022

    2021

  31. [32]

    Deep learning-based prediction of Monte Carlo dose distribution for heavy ion therapy

    R. He, H. Zhang, J. Wang, et al. “Deep learning-based prediction of Monte Carlo dose distribution for heavy ion therapy”. In:Physics and Imaging in Radiation Oncology34 (2025), p. 100735

    2025

  32. [33]

    MCDNet–a denoising convolutional neural network to accelerate Monte Carlo radiation transport simulations: A proof of principle with patient dose from x-ray CT imaging

    Z. Peng, H. Shan, T. Liu, et al. “MCDNet–a denoising convolutional neural network to accelerate Monte Carlo radiation transport simulations: A proof of principle with patient dose from x-ray CT imaging”. In:Ieee Access7 (2019), pp. 76680–76689

    2019

  33. [34]

    A novel multichannel deep learning model for fast denoising of Monte Carlo dose calculations: preclinical applica- tions

    R. H. van Dijk, N. Staut, C. J. Wolfs, et al. “A novel multichannel deep learning model for fast denoising of Monte Carlo dose calculations: preclinical applica- tions”. In:Physics in Medicine & Biology67.16 (2022), p. 164001

    2022

  34. [35]

    Accurate and versatile 3D segmentation of plant tissues at cellular resolution

    A. Wolny, L. Cerrone, A. Vijayan, et al. “Accurate and versatile 3D segmentation of plant tissues at cellular resolution”. In:Elife9 (2020), e57613

    2020

  35. [36]

    Superhuman accuracy on the SNEMI3D connectomics challenge

    K. Lee, J. Zung, P. Li, et al. “Superhuman accuracy on the SNEMI3D connectomics challenge”. In:arXiv preprint arXiv:1706.00120(2017)

    work page arXiv 2017

  36. [37]

    A plan verification platform for online adaptive proton therapy using deep learning-based Monte–Carlo denoising

    G. Zhang, X. Chen, J. Dai, et al. “A plan verification platform for online adaptive proton therapy using deep learning-based Monte–Carlo denoising”. In:Physica Medica103 (2022), pp. 18–25

    2022

  37. [38]

    AnatomyNet: deep learning for fast and fully automated whole-volume segmentation of head and neck anatomy

    W. Zhu, Y . Huang, L. Zeng, et al. “AnatomyNet: deep learning for fast and fully automated whole-volume segmentation of head and neck anatomy”. In:Medical physics46.2 (2019), pp. 576–589

    2019

  38. [39]

    Squeeze-and-excitation networks

    J. Hu, L. Shen, and G. Sun. “Squeeze-and-excitation networks”. In:Proceedings of the IEEE conference on computer vision and pattern recognition. 2018, pp. 7132–7141

    2018

  39. [40]

    Swin transformer: Hier- archical vision transformer using shifted windows

    Z. Liu, Y . Lin, Y . Cao, et al. “Swin transformer: Hier- archical vision transformer using shifted windows”. In: Proceedings of the IEEE/CVF international conference on computer vision. 2021, pp. 10012–10022

    2021

  40. [41]

    TransUNet: Transformers Make Strong Encoders for Medical Image Segmentation

    J. Chen, Y . Lu, Q. Yu, et al. “Transunet: Transformers make strong encoders for medical image segmentation”. In:arXiv preprint arXiv:2102.04306(2021)

    work page internal anchor Pith review arXiv 2021

  41. [42]

    3d transunet: Advancing medical image segmentation through vision transform- ers. arXiv 2023

    J. Chen, J. Mei, X. Li, et al. “3d transunet: Advancing medical image segmentation through vision transform- ers. arXiv 2023”. In:arXiv preprint arXiv:2310.07781 (2023)

    work page arXiv 2023

  42. [43]

    Swin-unet: Unet- like pure transformer for medical image segmentation

    H. Cao, Y . Wang, J. Chen, et al. “Swin-unet: Unet- like pure transformer for medical image segmentation”. In:European conference on computer vision. Springer. 2022, pp. 205–218

    2022

  43. [44]

    SegFormer: Simple and efficient design for semantic segmentation with trans- formers

    E. Xie, W. Wang, Z. Yu, et al. “SegFormer: Simple and efficient design for semantic segmentation with trans- formers”. In:Advances in neural information processing systems34 (2021), pp. 12077–12090

    2021

  44. [45]

    Mitigating inherent noise in Monte Carlo dose distributions us- ing dilated U-Net

    U. Javaid, K. Souris, D. Dasnoy, et al. “Mitigating inherent noise in Monte Carlo dose distributions us- ing dilated U-Net”. In:Medical Physics46.12 (2019), pp. 5790–5798

    2019

  45. [46]

    Single shot full plan deep learning dose computation for radiation therapy using spherical harmonics

    M. F. Kraus, R. Gao, S. Arberet, et al. “Single shot full plan deep learning dose computation for radiation therapy using spherical harmonics”. In:Medical Physics 53.1 (2026), e70223

    2026

  46. [47]

    Differentiable Gamma Index-Based Loss Functions: Accelerating Monte-Carlo Radiotherapy Dose Simula- tion

    S. Martinot, N. Komodakis, M. Vakalopoulou, et al. “Differentiable Gamma Index-Based Loss Functions: Accelerating Monte-Carlo Radiotherapy Dose Simula- tion”. In:International Conference on Information Pro- cessing in Medical Imaging. Springer. 2023, pp. 485– 496

    2023

  47. [48]

    Project monai

    M. Consortium et al. “Project monai”. In:Zenodo (2020)

    2020

  48. [49]

    Tomotherapy treatment plan quality assurance: the impact of applied criteria on passing rate in gamma index method

    S. Bresciani, A. D. Dia, A. Maggio, et al. “Tomotherapy treatment plan quality assurance: the impact of applied criteria on passing rate in gamma index method”. In: Medical physics40.12 (2013), p. 121711

    2013

  49. [50]

    Gamma anal- ysis dependence on specified low-dose thresholds for VMAT QA

    J.-H. Song, M.-J. Kim, S.-H. Park, et al. “Gamma anal- ysis dependence on specified low-dose thresholds for VMAT QA”. In:Journal of Applied Clinical Medical Physics16.6 (2015), pp. 263–272

    2015

  50. [51]

    To- talSegmentator: robust segmentation of 104 anatomic structures in CT images

    J. Wasserthal, H.-C. Breit, M. T. Meyer, et al. “To- talSegmentator: robust segmentation of 104 anatomic structures in CT images”. In:Radiology: Artificial In- telligence5.5 (2023), e230024

    2023

  51. [52]

    PlatiPy: processing library and analysis toolkit for medical imaging in Python

    P. Chlap and R. N. Finnegan. “PlatiPy: processing library and analysis toolkit for medical imaging in Python”. In:Journal of Open Source Software8.86 (2023), p. 5374

    2023

  52. [53]

    Current and emerging trends in medical image seg- mentation with deep learning

    P.-H. Conze, G. Andrade-Miranda, V . K. Singh, et al. “Current and emerging trends in medical image seg- mentation with deep learning”. In:IEEE Transactions on Radiation and Plasma Medical Sciences7.6 (2023), pp. 545–569

    2023

  53. [54]

    A review on low-dose emission tomography post-reconstruction denoising with neural network approaches

    A. Bousse, V . S. S. Kandarpa, K. Shi, et al. “A review on low-dose emission tomography post-reconstruction denoising with neural network approaches”. In:IEEE transactions on radiation and plasma medical sciences 8.4 (2024), pp. 333–347

    2024