Pith. sign in

REVIEW 4 major objections 5 minor 51 references

Diffusion Transformer-based Universal Dose Denoising for Pencil Beam Scanning Proton Therapy

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A diffusion-transformer denoiser trained on head-and-neck plans reconstructs high-statistics Monte Carlo dose maps for lung, breast, and prostate cases with no retraining, keeping whole-body MAE below 0.4 Gy[RBE] and 3%/2mm gamma pass…

desk verdict The cross-site generalization claim is plausible but unproven: without a raw-input baseline or a simple-filter baseline, the reported MAEs and gamma rates could just reflect how close the 1-minute MC already is to the 10-minute MC. read the letter →

arxiv 2506.04467 v1 pith:XVWAO3OX submitted 2025-06-04 physics.med-ph cs.AI

Yuzhen Ding , Jason Holmes , Hongying Feng , Martin Bues , Lisa A. McGee , Jean-Claude M. Rwigema , Nathan Y. Yu , Terence S. Sio
show 8 more authors
Sameer R. Keole William W. Wong Steven E. Schild Jonathan B. Ashman Sujay A. Vora Daniel J. Ma Samir H. Patel Wei Liu
This is my paper
classification physics.med-phcs.AI
keywords diffusiontransformerdosedenoisingprotontherapyMonteCarlosimulationpencilbeamscanningonlineadaptiveradiationgeneralizationheadandneckcancer
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 tries to establish that a single dose-denoising model, trained only on head-and-neck pencil-beam proton therapy plans, can clean up low-statistics Monte Carlo dose maps from other disease sites without retraining. The input is a one-minute MCsquare dose map plus the planning CT; the output is meant to match a ten-minute, high-statistics dose map. On held-out head-and-neck, lung, breast, and prostate cases, whole-body mean absolute error stayed below 0.4 Gy[RBE] and 3%/2mm gamma pass rates stayed above 92%. If the claim holds, online adaptive proton therapy could replace slow high-statistics Monte Carlo recalculations with a fast one-minute simulation followed by this denoiser, and one model would serve multiple anatomical sites.

What carries the argument

The load-bearing object is the diffusion transformer (DiT) backbone used as a conditional denoiser: instead of generating natural images in a latent space, the model operates directly on chunked dose/CT tensors of shape $4 \times 32 \times 32$ and learns the reverse-diffusion mapping $p_\theta(x_{t-1} \mid x_t, x_{\text{noise}}, y_{\text{ct}})$ with eight transformer blocks. The paired preprocessing pipeline is equally central: every 3D volume is flattened, cut into non-overlapping $1 \times 4096$ chunks, zero-padded, normalized with a log1p transform, and reconstructed by spatial concatenation, while CT Hounsfield units are non-linearly mapped to emphasize soft-tissue values around $-200$ to $300$ HU. The training objective combines the simplified diffusion noise-prediction loss $\mathbb{E}\|\epsilon - \epsilon_\theta\|^2$ with a weighted MAE on high-dose voxels and a residual loss that targets the top 10% and bottom 10% dose values, which is what steers the model toward clinically relevant target and fall-off regions.

What would settle it

Compute the 3%/2mm gamma pass rate and MAE separately for voxels lying at chunk boundaries (the first or last slice of each 4096-voxel chunk) versus voxels in chunk interiors across the 40 test cases; a systematic boundary-only degradation would show that the 1D chunking discards cross-chunk spatial context and would refute the claim that spatial context is fully preserved.

Watch

Extended reading notes

Core claim

The central claim is that a diffusion-transformer denoiser conditioned on the noisy dose map and the planning CT recovers the high-statistics dose map without any site-specific retraining. The authors build this from the standard denoising diffusion framework: the reverse process predicts the added noise at each diffusion step given the noisy dose, the CT, and the current noisy latent, with training loss combining the diffusion simple loss, a weighted mean absolute error, and a residual loss on the top and bottom 10% dose voxels. After a shared preprocessing pipeline that flattens volumes into 1x4096 chunks, zero-pads, normalizes dose by log1p, and non-linearly maps CT Hounsfield units, the model is trained on 80 head-and-neck patients and tested on 10 patients from each of four sites. Whole-body MAE was 0.195 Gy[RBE] for head-and-neck, 0.120 for lung, 0.172 for breast, and 0.376 for prostate, with 3D gamma pass rates above 92% (3%/2mm) at all sites and DVH indices in close agreement.

Load-bearing premise

The method assumes that flattening each 3D dose and CT volume into a 1D vector, cutting it into non-overlapping 1x4096 chunks, zero-padding, and then stitching the denoised chunks back together preserves enough spatial context for voxel-accurate denoising, especially across chunk boundaries.

Editorial extensions

If this is right

  • A single head-and-neck-trained model can be applied to lung, breast, and prostate cases with no fine-tuning, reducing the need for site-specific dose denoising models.
  • Online adaptive proton therapy workflows could substitute a roughly one-minute low-statistics MCsquare run plus denoising for a roughly ten-minute high-statistics run while keeping whole-body gamma pass rates above 92% at 3%/2mm.
  • Because CTV and OAR DVH indices track the ground truth, the denoised dose maps are clinically usable for target coverage and organ-sparing assessment.
  • The authors note that for applications demanding very high precision, fine-tuning on a small task-specific dataset or aligning preprocessing to the training distribution is advisable.

Reading between the lines

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

  • An implicit testable extension is whether the same chunk-based preprocessing transfers to other dose engines or to higher-noise regimes such as 30-second MC runs, which would determine the practical speed-gain ceiling.
  • The 1D chunking may make the model insensitive to long-range anatomical context; a natural experiment is to compare boundary voxels versus interior voxels on gamma analysis, since a mismatch would indicate the denoiser is exploiting local statistics rather than true 3D dose transport.
  • Because the model was trained without structure contours, it treats all non-zero dose voxels equally; adding contour or beam-geometry conditioning would likely raise CTV-specific gamma pass rates, which the authors themselves identify as a future step.
  • The prostate whole-body MAE is the largest of the four sites and has a wide spread, so a site-stratified analysis with more prostate cases would clarify whether the universal model degrades systematically with body size or pelvic bone heterogeneity.
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 / 5 minor

Summary. This manuscript proposes a diffusion-transformer-based universal dose denoising framework for pencil-beam scanning proton therapy. The authors generate noisy and high-statistics dose maps with MCsquare (about 1 min and 10 min per plan), train a conditional diffusion transformer on 80 H&N patients using noisy dose and CT images as inputs and high-statistics dose as ground truth, and test on held-out H&N, lung, breast, and prostate patients. They report whole-body MAE between 0.120 and 0.376 Gy[RBE], 3D gamma pass rates above 92% at 3%/2mm, and close DVH agreement, and conclude that an H&N-only-trained model generalizes to other disease sites without fine-tuning.

Significance. If the central claim holds, a single H&N-trained denoiser that works across disease sites without contours, fixed grid sizes, or retraining would be practically valuable for online adaptive proton therapy. The study has real strengths: evaluation on held-out patients from four disease sites, use of the open-source MCsquare engine, inclusion of clinically relevant DVH indices, and inference that requires no structure contours. However, the absence of any metric for the raw 1-minute noisy input against the same ground truth means the reported numbers do not yet establish that the network improves on the input; they may simply quantify MC noise. The spatial-context argument for the 1D chunking is also unvalidated. These gaps are fixable and should be addressed before the universal-denoising claim can be accepted.

major comments (4)
  1. [Abstract and Results (Dose Distribution Comparison)] The core claim of the paper is that the model denoises low-statistics MC dose maps, but the evaluation only compares the denoised output to the 10-minute ground truth. No MAE, gamma pass rate, or DVH metric is reported for the raw 1-minute MCsquare noisy input against the same ground truth. Without this control, the whole-body MAEs of 0.120-0.376 Gy[RBE] and gamma pass rates above 92% could be explained by the input already being dosimetrically close to the ground truth, with the network acting as a near-identity map. Please add, for each disease site, the raw noisy-input MAE and 3D gamma pass rates (and ideally a Gaussian-filtered or local-mean baseline) and report paired differences with confidence intervals between raw input and denoised output.
  2. [Methods (Data Collection and Processing), steps 1-2 and 7] The preprocessing flattens the 3D dose and CT volumes into 1D vectors, cuts them into non-overlapping 1x4096 chunks, zero-pads, and reconstructs the output by concatenation. This discards 3D neighborhood structure across chunk boundaries and across slice boundaries, and the binary CT mask 'five surrounding voxels' in step 4 is defined in the flattened 1D representation rather than in 3D. The manuscript does not provide an ablation or a comparison with a 3D patch-based or volume-based model, so it is unclear whether boundary artifacts at tissue interfaces or in the distal fall-off region are hidden by whole-body metrics. Please report metrics restricted to voxels near chunk boundaries and compare against a spatially local 3D model.
  3. [Results (Dose Distribution Comparison) and statistical analysis] The p-values reported throughout the results (e.g., p<.05 for whole-body MAE in H&N) are presented without a stated null hypothesis. If they come from a one-sample test of MAE against zero, they are uninformative because any positive MAE is trivially significant with enough voxels; if they compare denoised against raw input, that comparison is not described. In addition, the large standard deviations for CTV MAE (e.g., lung 1.08 +/- 2.03 Gy[RBE] and prostate 1.02 +/- 1.52 Gy[RBE]) suggest strong heterogeneity or outliers; please show patient-level distributions and median/IQR values in addition to means.
  4. [Abstract conclusion and Discussion] The abstract and conclusion state that the model can accurately and robustly denoise across disease sites, but no pre-specified quantitative acceptance criterion is given. The Discussion itself acknowledges that fine-tuning is advisable for high-precision applications and that CTV gamma pass rates are not consistently higher than whole-body rates, which tempers the universal claim. Please define a tolerance or clinical criterion for 'accurate and robust' (e.g., relative to prescription dose or a gamma threshold) and report per-site failure rates, so the conclusion can be assessed against a fixed standard.
minor comments (5)
  1. [Methods, step 3] Step 3 says the noisy doses were 'smoothed with Gaussian noise N(0; I)', which would add noise rather than smooth it; Figure 1 and its caption describe a Gaussian filter. Please clarify which operation was actually applied and report the filter parameters.
  2. [Discussion vs Table 1] The Discussion states that HU normalization emphasizes the range -300 to 200, whereas Methods and Table 1 specify -200 to 300 HU; these should be aligned.
  3. [Abstract and Results] There are several formatting and typographical issues, including 'approximately 1 minutes' in the Abstract, missing leading zeros in values such as '.120 +/- .054', and 'presentative' instead of 'representative' in the Results.
  4. [Results, prostate OAR] The prostate OAR MAE is reported as '.02 +/- .02 Gy[RBE] and 1.72 +/- .09 %'; the large discrepancy between the relative error and the tiny absolute value suggests a transcription or unit error, and should be checked.
  5. [Results, p-values] The prostate whole-body result is marked p<.005 while all other sites are marked p<.05; please state the actual p-values and the test used, rather than threshold symbols.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: held-out test evaluation and external MCsquare ground truths make the reported metrics independent of the training objective.

full rationale

The paper's central claim is that an H&N-trained diffusion transformer denoises 1-minute MCsquare dose maps toward 10-minute high-statistics references across four disease sites. The reported MAE, 3D gamma, and DVH metrics are computed on held-out test patients, so they are not fitted on the test set and do not reduce to the training loss by construction. The training labels and evaluation references are both produced by MCsquare, but at different particle statistics; this is a standard supervised denoising setup rather than a circular definition. Self-citations to prior Mayo Clinic work appear in the introduction and methods, but they are not load-bearing for the central result: MCsquare is an external open-source code, the DiT backbone is cited to Peebles et al., and the experimental evaluation uses independent test data. The most significant limitation—that no raw 1-minute noisy-dose-to-ground-truth baseline is reported—is a missing control for demonstrating improvement over the input, not a circularity. Without comparing the noisy input to ground truth, one cannot tell how much of the reported MAE reflects denoising versus the input noise level, but this is an experimental-design gap, not a self-referential reduction. No equation, fitted parameter, or citation chain is equivalent to the claimed prediction by construction, so no circular step is identified.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The central claim rests on the MCsquare high-statistics reference, the chunking representation, the Gaussian filter, and the small held-out cohorts. No physical entities are invented; all assumptions are data and modeling choices. The model does not derive any first-principles result, so the ledger mostly measures unvalidated preprocessing decisions.

free parameters (6)
  • Number of DiT blocks = 8
    Set to 8 to avoid gradient vanishing and overfitting (Methods, model architecture improvements, item 3); no ablation or sensitivity analysis is reported.
  • Chunk length and shape = 1x4096 reshaped to 4x32x32
    Each 3D volume is vectorized and cut into non-overlapping chunks of 4096 voxels (Methods step 2); this defines the receptive field and is chosen by hand without justification or ablation.
  • CT mask dilation radius = 5 surrounding voxels
    Binary CT mask includes non-zero dose voxels plus five surrounding voxels (Methods step 4); no sensitivity analysis is provided.
  • Residual loss dose quantiles = top 10% and bottom 10%
    The residual loss targets voxels with top and bottom 10% dose values (Methods, loss definition); chosen by hand and not varied.
  • HU normalization breakpoints = -200, 300, 3000, 29000 HU (Table 1)
    The nonlinear CT mapping emphasizes soft tissue and tumor HU range; breakpoints are selected by hand and not tested for sensitivity.
  • Gaussian filter parameters for noisy dose
    The preprocessing applies a Gaussian filter to the noisy dose (Figure 1), but the kernel size and sigma are not given; this preprocessing may itself perform much of the denoising and is never evaluated as a baseline.
assumptions (5)
  • domain assumption The 10-minute MCsquare high-statistics dose is an accurate clinical ground truth.
    Used as the supervised label and evaluation reference for all patients; no independent measurement, second Monte Carlo engine, or phantom validation is provided.
  • ad hoc to paper Flattening 3D volumes into non-overlapping 1D chunks preserves the spatial context required for dose denoising.
    Methods steps 1-2 discard inter-chunk 3D context; the paper asserts spatial relationships are preserved but provides no ablation or boundary artifact analysis.
  • domain assumption The Gaussian filter removes Monte Carlo noise without removing clinically relevant dose structure.
    Figure 1 states a Gaussian filter is applied exclusively to the noisy dose; if it suppresses real high-frequency dose features, the network cannot recover them.
  • domain assumption Ten test patients per disease site are representative of the broader population for each site.
    The cross-site generalization claim is based on 10 lung, 10 breast, and 10 prostate cases from a single institution; no external or multi-institutional validation is included.
  • domain assumption Monte Carlo noise in low-statistics proton dose maps is sufficiently close to the Gaussian noise assumed by the diffusion forward process.
    DDPM training (Equations 1-4) assumes Gaussian corruption; low-statistics MC noise in zero-inflated, low-dose regions is not Gaussian, and the paper does not characterize the noise distribution.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Diffusion Transformer-based Universal Dose Denoising for Pencil Beam Scanning Proton Therapy." pith.science (2026). https://pith.science/paper/XVWAO3OX

@misc{pith2026250604467,
  author       = {Pith},
  title        = {Pith review of: Diffusion Transformer-based Universal Dose Denoising for Pencil Beam Scanning Proton Therapy},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XVWAO3OX}},
  note         = {Machine review of arXiv:2506.04467}
}
read the original abstract

Purpose: Intensity-modulated proton therapy (IMPT) offers precise tumor coverage while sparing organs at risk (OARs) in head and neck (H&N) cancer. However, its sensitivity to anatomical changes requires frequent adaptation through online adaptive radiation therapy (oART), which depends on fast, accurate dose calculation via Monte Carlo (MC) simulations. Reducing particle count accelerates MC but degrades accuracy. To address this, denoising low-statistics MC dose maps is proposed to enable fast, high-quality dose generation. Methods: We developed a diffusion transformer-based denoising framework. IMPT plans and 3D CT images from 80 H&N patients were used to generate noisy and high-statistics dose maps using MCsquare (1 min and 10 min per plan, respectively). Data were standardized into uniform chunks with zero-padding, normalized, and transformed into quasi-Gaussian distributions. Testing was done on 10 H&N, 10 lung, 10 breast, and 10 prostate cancer cases, preprocessed identically. The model was trained with noisy dose maps and CT images as input and high-statistics dose maps as ground truth, using a combined loss of mean square error (MSE), residual loss, and regional MAE (focusing on top/bottom 10% dose voxels). Performance was assessed via MAE, 3D Gamma passing rate, and DVH indices. Results: The model achieved MAEs of 0.195 (H&N), 0.120 (lung), 0.172 (breast), and 0.376 Gy[RBE] (prostate). 3D Gamma passing rates exceeded 92% (3%/2mm) across all sites. DVH indices for clinical target volumes (CTVs) and OARs closely matched the ground truth. Conclusion: A diffusion transformer-based denoising framework was developed and, though trained only on H&N data, generalizes well across multiple disease sites.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

51 extracted references · 45 canonical work pages

  1. [1]

    Proton therapy–present and future

    Mohan R, Grosshans D. Proton therapy–present and future. Advanced drug delivery reviews. 2017;109:26-44

    work page 2017

  2. [2]

    A model- based approach to predict short-term toxicity benefits with proton therapy for oropharyngeal cancer

    Rwigema J-CM, Langendijk JA, van der Laan HP , Lukens JN, Swisher-McClure SD, Lin A. A model- based approach to predict short-term toxicity benefits with proton therapy for oropharyngeal cancer. International Journal of Radiation Oncology* Biology* Physics. 2019;104(3):553-62

    work page 2019

  3. [3]

    Intensity-modulated proton beam therapy (IMPT) versus intensity-modulated photon therapy (IMRT) for patients with oropharynx cancer–a case matched analysis

    Blanchard P , Garden AS, Gunn GB, Rosenthal DI, Morrison WH, Hernandez M, Crutison J, Lee JJ, Ye R, Fuller CD. Intensity-modulated proton beam therapy (IMPT) versus intensity-modulated photon therapy (IMRT) for patients with oropharynx cancer–a case matched analysis. Radiotherapy and Oncology. 2016;120(1):48-55

    work page 2016

  4. [4]

    Proton beam therapy for locally advanced lung cancer: A review

    Schild SE, Rule WG, Ashman JB, Vora SA, Keole S, Anand A, Liu W, Bues M. Proton beam therapy for locally advanced lung cancer: A review. World J Clin Oncol. 2014;5(4):568-75. Epub 2014/10/11. doi: 10.5306/wjco.v5.i4.568. PubMed PMID: 25302161; PMCID: PMC4129522

    doi:10.5306/wjco.v5.i4.568 2014

  5. [5]

    Robust intensity-modulated proton therapy to reduce high linear energy transfer in organs at risk

    An Y , Shan J, Patel SH, Wong W, Schild SE, Ding X, Bues M, Liu W. Robust intensity-modulated proton therapy to reduce high linear energy transfer in organs at risk. Med Phys. 2017;44(12):6138-47. Epub 2017/10/05. doi: 10.1002/mp.12610. PubMed PMID: 28976574; PMCID: PMC5734644

    work page doi:10.1002/mp.12610 2017

  6. [6]

    Matney J, Park PC, Bluett J, Chen YP , Liu W, Court LE, Liao Z, Li H, Mohan R. Effects of respiratory motion on passively scattered proton therapy versus intensity modulated photon therapy for stage III lung cancer: are proton plans more sensitive to breathing motion? International journal of radiation oncology, biology, physics. 2013;87(3):576-82. Epub 2...

    doi:10.1016/j.ijrobp.2013.07.007 2013

  7. [7]

    Parameterization of multiple Bragg curves for scanning proton beams using simultaneous fitting of multiple curves

    Zhang X, Liu W , Li Y , Li X, Quan M, Mohan R, Anand A, Sahoo N, Gillin M, Zhu XR. Parameterization of multiple Bragg curves for scanning proton beams using simultaneous fitting of multiple curves. Phys Med Biol. 2011;56(24):7725-35. Epub 2011/11/17. doi: 10.1088/0031- 9155/56/24/003. PubMed PMID: 22085829

    doi:10.1088/0031- 2011

  8. [8]

    Robust optimization for spot-scanning proton therapy based on dose-linear-energy-transfer volume constraints

    Chen J, Yang Y , Feng H, Zhang L, Liu Z, Liu T, Vargas CE, Yu NY , Rwigema J-CM, Keole SR. Robust optimization for spot-scanning proton therapy based on dose-linear-energy-transfer volume constraints. International Journal of Radiation Oncology* Biology* Physics. 2025;121(5):1303-15

    work page 2025

Show all 51 references
  1. [9]

    Intensity modulated proton therapy and its sensitivity to treatment uncertainties 1: the potential effects of calculational uncertainties

    Lomax A. Intensity modulated proton therapy and its sensitivity to treatment uncertainties 1: the potential effects of calculational uncertainties. Physics in Medicine & Biology. 2008;53(4):1027

    2008

  2. [10]

    The calibration of CT Hounsfield units for radiotherapy treatment planning

    Schneider U, Pedroni E, Lomax A. The calibration of CT Hounsfield units for radiotherapy treatment planning. Physics in Medicine & Biology. 1996;41(1):111

    1996

  3. [11]

    The precision of proton range calculations in proton radiotherapy treatment planning: experimental verification of the relation between CT-HU and proton stopping power

    Schaffner B, Pedroni E. The precision of proton range calculations in proton radiotherapy treatment planning: experimental verification of the relation between CT-HU and proton stopping power. Physics in Medicine & Biology. 1998;43(6):1579

    1998

  4. [12]

    Dosimetric benefits of robust treatment planning for intensity modulated proton therapy for base-of- skull cancers

    Liu W, Mohan R, Park P , Liu Z, Li H, Li X, Li Y , Wu R, Sahoo N, Dong L, Zhu XR, Grosshans DR. Dosimetric benefits of robust treatment planning for intensity modulated proton therapy for base-of- skull cancers. Pract Radiat Oncol. 2014;4(6):384-91. Epub 2014/11/20. doi: 10.10...

    2014 doi

  5. [13]

    Exploratory study of the association of volumetric modulated arc therapy (VMAT) plan robustness with local failure in head and neck cancer

    Liu W, Patel SH, Harrington DP , Hu Y , Ding X, Shen J, Halyard MY , Schild SE, Wong WW, Ezzell GE, Bues M. Exploratory study of the association of volumetric modulated arc therapy (VMAT) plan robustness with local failure in head and neck cancer. J Appl Clin Med Phys. 2017;18...

    2017 doi

  6. [14]

    Robustness quantification methods comparison in volumetric modulated arc therapy to treat head and neck cancer

    Liu W, Patel SH, Shen JJ, Hu Y , Harrington DP , Ding X, Halyard MY , Schild SE, Wong WW, Ezzell GA, Bues M. Robustness quantification methods comparison in volumetric modulated arc therapy to treat head and neck cancer. Pract Radiat Oncol. 2016;6(6):e269-e75. Epub 2016/03/31....

    2016 doi

  7. [15]

    Robust optimization in intensity-modulated proton therapy

    Liu W, Zhang X, Li Y , Mohan R. Robust optimization in intensity-modulated proton therapy. Med Phys. 2012;39:1079-91

    2012

  8. [16]

    Robust optimization in IMPT using quadratic objective functions to account for the minimum MU constraint

    Shan J, An Y , Bues M, Schild SE, Liu W. Robust optimization in IMPT using quadratic objective functions to account for the minimum MU constraint. Med Phys. 2018;45(1):460-9. Epub 2017/11/18. doi: 10.1002/mp.12677. PubMed PMID: 29148570; PMCID: PMC5774242

    2018 doi

  9. [17]

    A novel and individualized robust optimization method using normalized dose interval volume constraints (NDIVC) for intensity-modulated proton radiotherapy

    Shan J, Sio TT, Liu C, Schild SE, Bues M, Liu W. A novel and individualized robust optimization method using normalized dose interval volume constraints (NDIVC) for intensity-modulated proton radiotherapy. Med Phys. 2019;46(1):382-93. Epub 2018/11/06. doi: 10.1002/mp.13276. Pu...

    2019 doi

  10. [18]

    Shan J, Yang Y , Schild SE, Daniels TB, Wong WW, Fatyga M, Bues M, Sio TT, Liu W. Intensity- modulated proton therapy (IMPT) interplay effect evaluation of asymmetric breathing with simultaneous uncertainty considerations in patients with non-small cell lung cancer. Med Phys. ...

    2020 doi

  11. [19]

    Worst case optimization: a method to account for uncertainties in the optimization of intensity modulated proton therapy

    Pflugfelder D, Wilkens J, Oelfke U. Worst case optimization: a method to account for uncertainties in the optimization of intensity modulated proton therapy. Physics in Medicine & Biology. 2008;53(6):1689

    2008

  12. [20]

    Accounting for range uncertainties in the optimization of intensity modulated proton therapy

    Unkelbach J, Chan TC, Bortfeld T. Accounting for range uncertainties in the optimization of intensity modulated proton therapy. Physics in Medicine & Biology. 2007;52(10):2755

    2007

  13. [21]

    Dosimetric comparison of distal esophageal carcinoma plans for patients treated with small-spot intensity-modulated proton versus volumetric-modulated arc therapies

    Liu C, Bhangoo RS, Sio TT, Yu NY , Shan J, Chiang JS, Ding JX, Rule WG, Korte S, Lara P , Ding X, Bues M, Hu Y , DeWees T, Ashman JB, Liu W. Dosimetric comparison of distal esophageal carcinoma plans for patients treated with small-spot intensity-modulated proton versus volume...

    2019 doi

  14. [22]

    Robust Optimization for Intensity Modulated Proton Therapy to Redistribute High Linear Energy Transfer from Nearby Critical Organs to Tumors in Head and Neck Cancer

    Liu C, Patel SH, Shan J, Schild SE, Vargas CE, Wong WW , Ding X, Bues M, Liu W. Robust Optimization for Intensity Modulated Proton Therapy to Redistribute High Linear Energy Transfer from Nearby Critical Organs to Tumors in Head and Neck Cancer. International journal of radiat...

    2020 doi

  15. [23]

    Impact of Spot Size and Spacing on the Quality of Robustly Optimized Intensity Modulated Proton Therapy Plans for Lung Cancer

    Liu C, Schild SE, Chang JY , Liao Z, Korte S, Shen J, Ding X, Hu Y , Kang Y , Keole SR, Sio TT, Wong WW, Sahoo N, Bues M, Liu W. Impact of Spot Size and Spacing on the Quality of Robustly Optimized Intensity Modulated Proton Therapy Plans for Lung Cancer. International journal...

    2018 doi

  16. [24]

    Small‐spot intensity‐modulated proton therapy and volumetric‐modulated arc therapies for patients with locally advanced non‐small‐cell lung cancer: a dosimetric comparative study

    Liu C, Sio TT, Deng W , Shan J, Daniels TB, Rule WG, Lara PR, Korte SM, Shen J, Ding X. Small‐spot intensity‐modulated proton therapy and volumetric‐modulated arc therapies for patients with locally advanced non‐small‐cell lung cancer: a dosimetric comparative study. Journal o...

    2018

  17. [25]

    Liu C, Yu NY , Shan J, Bhangoo RS, Daniels TB, Chiang JS, Ding X, Lara P , Patrick CL, Archuleta JP , DeWees T, Hu Y , Schild SE, Bues M, Sio TT, Liu W. Technical Note: Treatment planning system (TPS) approximations matter - comparing intensity-modulated proton therapy (IMPT) ...

    2019 doi

  18. [26]

    Per-voxel constraints to minimize hot spots in linear energy transfer-guided robust optimization for base of skull head and neck cancer patients in IMPT

    Feng H, Shan J, Anderson JD, Wong WW, Schild SE, Foote RL, Patrick CL, Tinnon KB, Fatyga M, Bues M, Patel SH, Liu W. Per-voxel constraints to minimize hot spots in linear energy transfer-guided robust optimization for base of skull head and neck cancer patients in IMPT. Med Ph...

    2022

  19. [27]

    Technical Note: 4D robust optimization in small spot intensity-modulated proton therapy (IMPT) for distal esophageal carcinoma

    Feng H, Shan J, Ashman JB, Rule WG, Bhangoo RS, Yu NY , Chiang J, Fatyga M, Wong WW, Schild SE, Sio TT, Liu W. Technical Note: 4D robust optimization in small spot intensity-modulated proton therapy (IMPT) for distal esophageal carcinoma. Med Phys. 2021;48(8):4636-47. Epub 202...

    2021 doi

  20. [28]

    Beam angle comparison for distal esophageal carcinoma patients treated with intensity-modulated proton therapy

    Feng H, Sio TT, Rule WG, Bhangoo RS, Lara P , Patrick CL, Korte S, Fatyga M, Wong WW, Schild SE, Ashman JB, Liu W. Beam angle comparison for distal esophageal carcinoma patients treated with intensity-modulated proton therapy. J Appl Clin Med Phys. 2020;21(11):141-52. Epub 202...

    2020 doi

  21. [29]

    GPU-accelerated Monte Carlo-based online adaptive proton therapy: A feasibility study

    Feng H, Patel SH, Wong WW, Younkin JE, Penoncello GP , Morales DH, Stoker JB, Robertson DG, Fatyga M, Bues M, Schild SE, Foote RL, Liu W. GPU-accelerated Monte Carlo-based online adaptive proton therapy: A feasibility study. Med Phys. 2022;49(6):3550-63. Epub 2022/04/21. doi: ...

    2022 doi

  22. [30]

    Online Adaptive Proton Therapy Facilitated by Artificial Intelligence-based Auto Segmentation in Pencil Beam Scanning Proton Therapy: Prostate oAPT in PBSPT

    Feng H, Shan J, Vargas CE, Keole SR, Rwigema J-CM, Yu NY , Ding Y , Zhang L, Schild SE, Wong WW, Vora SA, Shen J, Liu W. Online Adaptive Proton Therapy Facilitated by Artificial Intelligence-based Auto Segmentation in Pencil Beam Scanning Proton Therapy: Prostate oAPT in PBSPT...

    2024 doi

  23. [31]

    Modelling small block aperture in an in-house developed GPU-accelerated Monte Carlo-based dose engine for pencil beam scanning proton therapy

    Feng H, Holmes JM, Vora SA, Stoker JB, Bues M, Wong WW, Sio TS, Foote RL, Patel SH, Shen J, Liu W. Modelling small block aperture in an in-house developed GPU-accelerated Monte Carlo-based dose engine for pencil beam scanning proton therapy. Phys Med Biol. 2024;69(3). Epub 202...

    2024 doi

  24. [32]

    Fast Monte Carlo dose calculation in proton therapy

    Holmes J, Feng H, Zhang L, Fix MK, Jiang SB, Liu W. Fast Monte Carlo dose calculation in proton therapy. Physics in Medicine & Biology. 2024;69(17):17TR01

    2024

  25. [33]

    Virtual particle monte carlo (VPMC), a new concept to avoid simulating secondary particles in proton therapy dose calculation

    Shan J, Feng H, Morales DH, Patel SH, Wong WW , Fatyga M, Bues M, Schild SE, Foote RL, Liu W. Virtual particle monte carlo (VPMC), a new concept to avoid simulating secondary particles in proton therapy dose calculation. Med Phys. 2022. Epub 2022/08/13. doi: 10.1002/mp.15913. ...

    2022 doi

  26. [34]

    MCsquare

    Deng W, Younkin JE, Souris K, Huang S, Augustine K, Fatyga M, Ding X, Cohilis M, Bues M, Shan J, Stoker J, Lin L, Shen J, Liu W. Technical Note: Integrating an open source Monte Carlo code "MCsquare" for clinical use in intensity-modulated proton therapy. Med Phys. 2020;47(6):...

    2020 doi

  27. [35]

    A pencil beam algorithm for proton dose calculations

    Hong L, Goitein M, Bucciolini M, Comiskey R, Gottschalk B, Rosenthal S, Serago C, Urie M. A pencil beam algorithm for proton dose calculations. Physics in Medicine & Biology. 1996;41(8):1305

    1996

  28. [36]

    Khan's the physics of radiation therapy: Lippincott Williams & Wilkins; 2014

    Khan FM, Gibbons JP . Khan's the physics of radiation therapy: Lippincott Williams & Wilkins; 2014

    2014

  29. [37]

    Fast pencil beam dose calculation for proton therapy using a double-Gaussian beam model

    Da Silva J, Ansorge R, Jena R. Fast pencil beam dose calculation for proton therapy using a double-Gaussian beam model. Frontiers in oncology. 2015;5:281

    2015

  30. [38]

    Clinical Validation of a Ray-Casting Analytical Dose Engine for Spot Scanning Proton Delivery Systems

    Younkin JE, Morales DH, Shen J, Shan J, Bues M, Lentz JM, Schild SE, Stoker JB, Ding X, Liu W. Clinical Validation of a Ray-Casting Analytical Dose Engine for Spot Scanning Proton Delivery Systems. Technol Cancer Res Treat. 2019;18:1533033819887182. Epub 2019/11/23. doi: 10.11...

    2019 doi

  31. [39]

    Deep-learning based fast and accurate 3D CT deformable image registration in lung cancer

    Ding Y , Feng H, Yang Y , Holmes J, Liu Z, Liu D, Wong WW, Yu NY , Sio TT, Schild SE, Li B, Liu W. Deep-learning based fast and accurate 3D CT deformable image registration in lung cancer. Med Phys. 2023;50(11):6864-80. Epub 20230608. doi: 10.1002/mp.16548. PubMed PMID: 372891...

    2023 doi

  32. [40]

    Accurate patient alignment without unnecessary imaging using patient-specific 3D CT images synthesized from 2D kV images

    Ding Y , Holmes JM, Feng H, Li B, McGee LA, Rwigema JM, Vora SA, Wong WW , Ma DJ, Foote RL, Patel SH, Liu W. Accurate patient alignment without unnecessary imaging using patient-specific 3D CT images synthesized from 2D kV images. Commun Med (Lond). 2024;4(1):241. Epub 2024112...

    2024 doi

  33. [41]

    Deep learning method for prediction of patient-specific dose distribution in breast cancer

    Ahn SH, Kim E, Kim C, Cheon W, Kim M, Lee SB, Lim YK, Kim H, Shin D, Kim DY . Deep learning method for prediction of patient-specific dose distribution in breast cancer. Radiation Oncology. 2021;16:1-13

    2021

  34. [42]

    DoseNet: An Ensemble-Based Deep Learning Method for 3D Dose Prediction in IMRT

    Wang Q, Song Y , Hu J, Liang L, editors. DoseNet: An Ensemble-Based Deep Learning Method for 3D Dose Prediction in IMRT. 2023 International Annual Conference on Complex Systems and Intelligent Science (CSIS-IAC); 2023: IEEE

    2023

  35. [43]

    DeepDoseNet: a deep learning model for 3D dose prediction in radiation therapy

    Soomro MH, Alves VGL, Nourzadeh H, Siebers JV. DeepDoseNet: a deep learning model for 3D dose prediction in radiation therapy. arXiv preprint arXiv:211100077. 2021

    2021

  36. [44]

    Deep learning–based dose prediction for automated, individualized quality assurance of head and neck radiation therapy plans

    Gronberg MP , Beadle BM, Garden AS, Skinner H, Gay S, Netherton T, Cao W, Cardenas CE, Chung C, Fuentes DT. Deep learning–based dose prediction for automated, individualized quality assurance of head and neck radiation therapy plans. Practical radiation oncology. 2023;13(3):e282-e91

    2023

  37. [45]

    A cascade transformer- based model for 3D dose distribution prediction in head and neck cancer radiotherapy

    Gheshlaghi T, Nabavi S, Shirzadikia S, Moghaddam ME, Rostampour N. A cascade transformer- based model for 3D dose distribution prediction in head and neck cancer radiotherapy. Physics in Medicine & Biology. 2024;69(4):045010

    2024

  38. [46]

    Scalable diffusion models with transformers

    Peebles W, Xie S, editors. Scalable diffusion models with transformers. Proceedings of the IEEE/CVF International Conference on Computer Vision; 2023

    2023

  39. [47]

    doi: 10.1002/mp.15384

    Epub 2021/11/30. doi: 10.1002/mp.15384. PubMed PMID: 34843119

    2021 doi

  40. [48]

    Deep unsupervised learning using nonequilibrium thermodynamics

    Sohl-Dickstein J, Weiss E, Maheswaranathan N, Ganguli S, editors. Deep unsupervised learning using nonequilibrium thermodynamics. International conference on machine learning; 2015: PMLR

    2015

  41. [49]

    Denoising diffusion probabilistic models

    Ho J, Jain A, Abbeel P . Denoising diffusion probabilistic models. Advances in neural information processing systems. 2020;33:6840-51

    2020

  42. [50]

    High-resolution image synthesis with latent diffusion models

    Rombach R, Blattmann A, Lorenz D, Esser P , Ommer B, editors. High-resolution image synthesis with latent diffusion models. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition; 2022

    2022

  43. [51]

    Auto-encoding variational bayes

    Kingma DP . Auto-encoding variational bayes. arXiv preprint arXiv:13126114. 2013

    2013

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.