Pith. sign in

REVIEW 3 major objections 5 minor 85 references

Mapping intratumoral heterogeneity through PET-derived washout and deep learning after proton therapy

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read This paper claims that an uncertainty-aware deep learning correction can recover voxel-wise washout rates and washed-out activity maps from offline post-proton-therapy PET, cutting median absolute errors by 60% and 28% and resolving…

desk verdict Solid in-silico proof of concept for DL-based voxel washout mapping, undercut by an abstract that overstates clinical readiness. read the letter →

arxiv 2506.21153 v1 pith:CPWQPWES submitted 2025-06-26 physics.med-ph

classification physics.med-ph
keywords biologicalwashoutpost-protontherapyPETdeeplearningintratumoralheterogeneityratemapswashed-outMonteCarlosimulationuncertaintyquantification
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

After proton therapy, the radioactive isotopes created inside the tumor can be imaged with PET to check where the dose landed, but the signal is degraded by biological washout: perfusion and metabolism carry the isotopes away, unevenly across the tumor. This paper tries to turn that nuisance into a biomarker. It trains an uncertainty-aware deep learning model on thousands of simulated "digital twin" tumors with known washout ingredients so that, from an ordinary offline PET scan taken 15 minutes after treatment, it can estimate voxel-by-voxel the slow washout rate ($\lambda_B$) and a new "washed-out" quantity ($A_0/A_{0,\mathrm{NW}}$), the fraction of activity that remains when imaging starts. In the paper's tests the model cuts median absolute errors by 60% for washout rate maps and 28% for washed-out maps relative to uncorrected exponential fits, with errors mostly below a reported threshold for telling vascularized from necrotic regions even in subregions as small as 5.1 mL. If the result holds in real patients, a single post-treatment PET scan would map intratumoral heterogeneity in vascularity and metabolism without any extra tracer or equipment, supporting dose verification, tumor characterization, and treatment adaptation.

What carries the argument

The carrying mechanism is the one-compartment kinetic model $A(t) = A_0 e^{-(\lambda_P+\lambda_B)t}$ for $^{11}\mathrm{C}$, justified by the observation that 15 minutes after treatment about 80% of the remaining activity comes from $^{11}\mathrm{C}$ and its fast and medium washout components have largely decayed. Two estimation routes hang on this equation: "fit correction" first fits the exponential to five 6-minute reconstructed PET frames voxel-by-voxel, producing the uncorrected $\lambda_B$ and $A_0$ maps, and then a self-attention U-Net trained with a variance-weighted negative log-likelihood plus structural-similarity loss corrects those maps; "direct estimation" lets the same network learn $\lambda_B$ maps straight from the PET frames. The paired noisy maps and ground truths that make this learning possible come from Monte Carlo simulation of proton therapy, PET acquisition, and PSF/TOF-aware MLEM reconstruction, with 75 synthetic tumor instances ("digital twins") per patient and irregular spheroid regions of assigned $\lambda_B$ and $A_0$ values.

What would settle it

Run a prospective patient study in which post-proton-therapy PET is acquired at the intended 15-minute delay and compare the deep-learning-corrected washout-rate and washed-out maps against an independent vascular or metabolic reference such as dynamic contrast-enhanced MRI or in-beam PET washout measurements in the same tumor; the central claim fails if the corrected maps cannot reproduce the reference's vascular pattern or if voxel errors on real data exceed the roughly 9e-3 $min^{-1}$ threshold used to claim 5 mL detectability.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that the washout information degraded in post-proton-therapy PET is recoverable by learning. A neural network, fed either uncorrected voxel-fit maps or raw reconstructed PET frames, outputs denoised and deblurred maps of the slow $^{11}\mathrm{C}$ washout rate $\lambda_B$ and of the washed-out ratio $A_0/A_{0,\mathrm{NW}}$. The fit-correction route is the more accurate one: median absolute error for $\lambda_B$ drops from $5.8\times10^{-3}\,\mathrm{min}^{-1}$ to $2.3\times10^{-3}\,\mathrm{min}^{-1}$ and median SSIM rises from 0.64 to 0.80, while washed-out maps improve from 2.9% to 2.1% median absolute error. The authors interpret these error levels as small enough to separate low-, medium-, and high-vascularity tumor subregions down to roughly 5 mL, and they report that the corrected model transfers to an unseen liver case and to a 10-minute training delay without losing accuracy. The result is presented as a proof that subtumoral washout kinetics can be mapped from routine offline PET, with well-calibrated uncertainty estimates that flag unreliable voxels.

Load-bearing premise

The load-bearing premise is that the Monte Carlo simulations, the five-frame reconstruction, and the simplified single-compartment exponential model reproduce how real tumors wash out isotopes, so if actual patient noise, isotope mixtures, or washout kinetics differ from these digital twins, the error reductions and 5 mL detectability threshold may not transfer to the clinic.

Editorial extensions

If this is right

  • A single 30-minute offline PET scan, with no extra radiotracer, could reveal subregions of tumor with different washout or vascularity down to about 5 mL, turning a standard verification scan into a heterogeneity assay.
  • The corrected washout maps can be combined with the authors' prior dose-verification workflow on the same PET data, so range verification and tumor characterization would come from one acquisition.
  • The reported transfer to a liver case and to a different acquisition delay suggests the method does not need retraining for every anatomical site or exact 15-minute schedule, though redevelopment may be needed for very different protocols.
  • Uncertainty thresholds can be used to discard unreliable voxels, improving classification accuracy, so clinical adoption could report both the map and a mask of where it is trustworthy.
  • By mapping slow washout rates, the framework could support dose painting or adaptive replanning that targets hypoxic or poorly perfused subregions, if the washout-vascularity link holds in humans.

Reading between the lines

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

  • Because the detectability threshold comes from a rat study using $^{15}\mathrm{O}$ beams, the paper's 5 mL claim implicitly assumes that human $^{11}\mathrm{C}$ slow-washout differences between vascular statuses are at least as large as the rat $^{15}\mathrm{O}$ differences; that assumption is testable but not yet demonstrated.
  • The washed-out map $A_0/A_{0,\mathrm{NW}}$ conflates the medium and fast component fractions with the isotope mix at imaging onset, so its biological meaning is not a single rate; if validated against perfusion imaging, it could act as a therapy-induced perfusion-like biomarker rather than another washout-rate map.
  • The direct-estimation route, though less accurate, removes the per-voxel curve fit and runs quickly (about half a second including the 20-pass uncertainty estimate), so it may become the practical choice for time-critical adaptation, whereas the fit-correction route is better for offline quantification.
  • Since the whole training set is simulated, the framework's real-world ceiling depends on how well the simulators capture scanner physics; a natural extension is to fine-tune the same architecture on a small set of real patient scans with follow-up outcomes, which the paper states is being pursued.
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

3 major / 5 minor

Summary. The paper introduces PROTOTWIN-PET Washout (PPW), a deep learning framework that estimates voxel-wise slow washout rates (lambda_B) and newly defined 'washed-out' maps (A0/A0,NW) from offline PET acquired 15 minutes after proton therapy. The authors generate 600 training and 300 test digital twins by combining patient CTs with FRED proton-therapy simulations and MCGPU-PET scanner simulations, introduce synthetic intratumoral heterogeneities with 1-5 irregular spheroid regions of distinct washout parameters, and compare two estimation routes: correction of voxel-wise fitted maps ('fit correction') and direct estimation from PET frames. In held-out simulated head-and-neck patients plus one simulated liver case, the fit-correction model reduces the median absolute error by 60% for lambda_B and 28% for washed-out maps relative to uncorrected fits, and the models generalize to a different acquisition delay. The authors propose these maps as surrogates for tumor vascular status and intratumoral heterogeneity, with uncertainty quantification via MC dropout and beta-NLL loss.

Significance. If the simulation pipeline faithfully reproduces the physics and biology of post-proton-therapy PET, the framework could convert routinely available offline PET into a non-invasive, tracer-free biomarker of tumor heterogeneity and vascular status. The study is methodologically solid: it ships open-source code, uses realistic Monte Carlo simulation, tests on held-out patients, reports paired Wilcoxon tests, and includes uncertainty quantification. Its significance, however, is currently that of an in silico proof of concept. Every accuracy metric is computed against ground-truth maps produced by the same simulation pipeline that generated the training data; no phantom or clinical data are used to test the physiological and physical assumptions. The clinical claims in the abstract and title therefore exceed the present evidence. The explicit acknowledgement in Section 4 that phantom and clinical validation is essential and ongoing is appropriate, but the framing of the results should be adjusted to match the simulation-only scope.

major comments (3)
  1. [Section 2.1, Eq. (2)] The reduction of the kinetic model to a single-exponential, pure-11C slow-washout component is load-bearing and insufficiently justified. At 15 minutes post-treatment, 13N (lambda_P=0.1003 min^-1, remaining fraction e^{-1.5045}=0.22) and 38K (lambda_P=0.1308 min^-1, remaining fraction 0.14) are still present, and the manuscript only states that about 80% of remaining activity is 11C without a derivation or a tissue-specific isotope-production calculation. Fitting Eq. (2) to a mixed-isotope signal will alias residual non-11C decay into lambda_B. Because the simulator, the ground-truth maps, and the training target all share this same simplified model, the network can learn to correct the specific bias of the assumed isotope mix, but this correction will not necessarily transfer to real tumors with different oxygen/nitrogen content or production cross-sections. I recommend adding a sensitivity analysis over isotope production ratios or a phantom experiment with a known isotope mixture to demonstrate that the learned correction is not tied to the particular simulated mix.
  2. [Section 4 and Figure 4a] The detectability claim stated in the abstract ('errors predominantly fell below thresholds for differentiating vascular status' for regions as small as 5 mL) is benchmarked against a washout-rate difference of 9x10^-3 min^-1 reported for 15O in rat tumors (reference [15]). The assumption that 11C washout-rate differences in human tumors are of similar magnitude is explicitly acknowledged in the Discussion but is not tested anywhere. In the absence of an independent phantom or clinical calibration, the threshold is not validated for 11C or for humans, and the claim should either be rephrased as 'errors fall below the threshold used in this simulation study' or be supported by a dedicated experiment with a known two-compartment ground truth.
  3. [Sections 2.1, step 4 and 2.3] The washed-out map A0/A0,NW is not a directly measurable clinical quantity. A0,NW is defined as the activity distribution that would be observed in the absence of biological washout, and the proposed inference workflow in Section 2.3 requires that this 'no-washout' map be re-estimated through a patient-specific proton-therapy and PET simulation. The clinical utility of the washed-out map is therefore conditional on the accuracy of the simulated isotope-production and scanner models, and it cannot be derived from the PET scan alone. This is a fundamental limitation for translation that should be stated prominently in the abstract and conclusions; currently the washed-out map is presented largely as a novel biomarker without emphasizing this simulation dependence.
minor comments (5)
  1. [Section 3.7, Table 4] The experiments with a 10-minute acquisition delay are reported in Table 4, but the Methods do not describe how the 10-minute dataset was generated (which steps of Section 2.1 were modified, whether the tumor heterogeneity distributions were identical, and how the shorter delay affects the isotope-mix assumption). Please add a short description to the Methods.
  2. [Section 2.4] The sentence '75 digital twins are generated for each patient, resulting in a total of 300 training cases' should read 'test cases' (or 'validation cases'), since these patients are used for evaluation, not training.
  3. [Section 2.1, step 4] The statement that approximately 80% of remaining activity originates from 11C after 15 minutes should be supported by a quantitative calculation or a specific reference that combines isotope production cross-sections and tissue composition; reference [45] is a cross-section library, but the 80% number needs a derivation.
  4. [Section 2.2.2, Eq. (3)] The notation with the floor brackets in the beta-NLL loss is unconventional; although the text explains the stop-gradient operation, the equation alone is ambiguous. A brief notational clarification would improve reproducibility.
  5. [Figure 2] In the schematic, the fit-correction branch is labeled with 'Uncorrected A0 and lambda_B maps' as input, but the same branch is also used for washed-out maps; the figure would be clearer if it explicitly indicated that the fit-correction branch can also process A0/A0,NW maps.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular reduction: washout-rate and washed-out targets are produced by an independent forward simulation chain, and same-group tool citations are code-reproduced rather than load-bearing.

full rationale

After walking the claimed derivation chain, I find no step in which a prediction is defined in terms of its own output or in which a fitted parameter is renamed as a prediction. The ground-truth lambda_B and A0/A0,NW maps are produced by a forward chain (FRED isotope production, Eq. 1/2 washout kinetics with independently assigned values, MCGPU-PET acquisition simulation, MLEM reconstruction) that is independent of the trained model; the model is then evaluated on held-out digital twins generated by the same chain. While this is a closed-loop simulation benchmark, the target maps are not computed from the model's estimates, so the error metrics in Tables 1-4 are not forced by construction. The same-group citations ([34] PROTOTWIN-PET and [50] MCGPU-PET) are code-reproduced, open-source tools and prior protocol work, not uniqueness theorems or ansatze, so they do not constitute load-bearing circularity. The paper explicitly states that 'Validation through phantom studies and clinical data is essential and is currently being pursued,' which identifies an external-validity gap rather than a circular derivation. The simulation does encode the authors' kinetic assumptions (e.g., Eq. 2 and the 15-minute, 11C-dominated delay), but that limits transfer to real patients, not the internal consistency of the claimed predictions.

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

The central results depend almost entirely on simulation assumptions: the kinetic model, the isotope simplification, the synthetic heterogeneity shapes, and the simulator fidelity. There are no fitted physical constants in the traditional sense, but the label generation and the interpretability threshold are set by the authors. The ledger is dominated by domain assumptions and ad hoc modeling choices that real-data validation would need to confirm.

free parameters (6)
  • Maximum slow washout rate (lambda_B,max) = 45 x 10^-3 min^-1
    Upper bound for ground truth washout rates, set slightly above reported human values (~35 x 10^-3 min^-1) to account for subregional variation and residual isotope contributions; defines the training label range and vascularity category thresholds.
  • Default washout component fractions (Ms, Mm, Mf) = 0.40, 0.30, 0.25 (varied by +-50%)
    Default fractions for slow, medium, and fast washout in Eq. 1, used to generate ground truth A0 and washed-out maps; taken from literature but selected by the authors for this simulation.
  • PET frame duration and count = 5 frames x 6 minutes
    Chosen as an optimal trade-off between number of time points and per-frame noise; directly shapes the inputs to both the kinetic fit and the deep learning model.
  • Number of digital twins per patient = 75
    Empirically determined: the authors report that adding more instances did not improve performance and fewer decreased it.
  • Model and loss hyperparameters = beta=0.25, SSIM weight=0.1, dropout=0.2, MC iterations=20, initial LR=1e-4
    Empirically optimized (stated in Section 2.2); these affect the reported error reduction and uncertainty calibration.
  • Voxel fit reliability threshold = 20% parameter uncertainty
    Voxels whose fitted parameters exceed 20% uncertainty are replaced by 3x3x3 neighborhood averages; this threshold is a heuristic choice affecting uncorrected map quality.
assumptions (5)
  • domain assumption Three-component biological washout model (Eq. 1) with fast, medium, and slow components.
    Adopted from prior literature (Tomitani 2003, Mizuno 2003) and used as the basis for all ground truth generation; the simulation assumes this model is correct.
  • domain assumption Single-isotope, single-component simplification A(t) = A0 exp(-(lambda_P + lambda_B) t) at 15 min delay.
    Assumes 11C contributes about 80% of activity and that fast/medium components have largely decayed by imaging onset; cited to Bauer et al. This equation is the target for washout rate estimation.
  • domain assumption Monte Carlo simulators (FRED for proton therapy, MCGPU-PET for PET acquisition) accurately reproduce real clinical physics.
    The paper provides no validation of the simulated PET against real patient scans; all training and testing rely on this toolchain.
  • ad hoc to paper Intratumoral heterogeneity can be represented by 1-5 irregular spheroids with uniform washout rates per region.
    A stylized ground truth; the authors acknowledge in Section 4 that the sharp boundaries are artificial because the simulator enforces fixed decay rates per region.
  • ad hoc to paper 11C washout kinetics in human tumors are comparable to 15O washout kinetics in rat tumors.
    The 9 x 10^-3 min^-1 threshold for differentiating vascular status comes from a rat 15O in-beam PET study (Toramatsu); the authors assume similarity to interpret the 5 mL detectability result.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Mapping intratumoral heterogeneity through PET-derived washout and deep learning after proton therapy." pith.science (2026). https://pith.science/paper/CPWQPWES

@misc{pith2026250621153,
  author       = {Pith},
  title        = {Pith review of: Mapping intratumoral heterogeneity through PET-derived washout and deep learning after proton therapy},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CPWQPWES}},
  note         = {Machine review of arXiv:2506.21153}
}
read the original abstract

The distribution of produced isotopes during proton therapy can be imaged with Positron Emission Tomography (PET) to verify dose delivery. However, biological washout, driven by tissue-dependent processes such as perfusion and cellular metabolism, reduces PET signal-to-noise ratio (SNR) and limits quantitative analysis. In this work, we propose an uncertainty-aware deep learning framework to improve the estimation of washout parameters in post-proton therapy PET, not only enabling accurate correction for washout effects, but also mapping intratumoral heterogeneity as a surrogate marker of tumor status and treatment response. We trained the models on Monte Carlo-simulated data from eight head-and-neck cancer patients, and tested them on four additional head-and-neck and one liver patient. Each patient was represented by 75 digital twins with distinct tumoral washout dynamics and imaged 15 minutes after treatment, when slow washout components dominate. We also introduced "washed-out" maps, quantifying the contribution of medium and fast washout components to the loss in activity between the end of treatment and the start of PET imaging. Trained models significantly improved resolution and accuracy, reducing average absolute errors by 60% and 28% for washout rate and washed-out maps, respectively. For intratumoral regions as small as 5 mL, errors predominantly fell below thresholds for differentiating vascular status, and the models generalized across anatomical areas and acquisition delays. This study shows the potential of deep learning in post-proton therapy PET to non-invasively map washout kinetics and reveal intratumoral heterogeneity, supporting dose verification, tumor characterization, and treatment personalization. The framework is available at https://github.com/pcabrales/ppw.git.

Figures

Figures reproduced from arXiv: 2506.21153 by the authors.

Figure 1
Figure 1. Start of imaging End of irradiation Physical Biological washout Decay Activity for isotope i Activity with No Washout ( ) [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Schematic overview of the PROTOTWIN-PET Washout framework, encompass [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Architecture of the nnFormer model, designed to output both the washout rate map [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Absolute error distributions obtained with the fit correction approach and without [PITH_FULL_IMAGE:figures/full_fig_p015_4.png]
Figure 5
Figure 5. Figure 5: Washout rate maps (λB) obtained with the fit correction approach (Corrected) and without correction (Uncorrected) are compared with the ground truth (GT) for three tumor samples in the test set, shown in a coronal view. The CT scan is included for anatomical context, w…
Figure 6
Figure 6. Figure 6: Washed-out maps (A0/A0,NW) obtained with the fit correction approach (Corrected) and without correction (Uncorrected) are compared with the ground truth (GT) for three tumor samples in the test set, shown in a coronal view. The CT scan is included for anatomical contex…
Figure 7
Figure 7. Figure 7: Washout rate (λB) sparsification error plot for all test set voxels. Errors are quantified using the median absolute error (MedAE). are of similar magnitude to those of 15O, such tissue-level differences could be detectable at subtumoral scales. Even in larger regions,…
Figure 8
Figure 8. Figure 8: Classification accuracy as a function of region volume when assigning each voxel [PITH_FULL_IMAGE:figures/full_fig_p019_8.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

85 extracted references · 73 canonical work pages

  1. [15]

    Toramatsu, A

    C. Toramatsu, A. Mohammadi, H. Wakizaka, H. Sudo, N. Nitta, C. Seki, I. Kanno, M. Takahashi, K. Karasawa, Y. Hirano, et al., Measurement of biological washout rates depending on tumor vascular status in 15O in- beam rat-PET, Physics in Medicine & Biology 67 (12) (2022) 125006

  2. [1]

    T. F. DeLaney, Proton therapy in the clinic, IMRT, IGRT, SBRT 43 (2011) 465–485

  3. [2]

    A. C. Kraan, A. Del Guerra, Technological developments and future per- spectives in particle therapy: A topical review, IEEE Transactions on Ra- diation and Plasma Medical Sciences (2024)

  4. [3]

    Paganetti, Range uncertainties in proton therapy and the role of Monte Carlo simulations, Physics in Medicine and Biology 57 (2012) R99 – R117

    H. Paganetti, Range uncertainties in proton therapy and the role of Monte Carlo simulations, Physics in Medicine and Biology 57 (2012) R99 – R117. doi:10.1088/0031-9155/57/11/R99

  5. [4]

    E. M. Quan, W. Liu, R. Wu, Y. Li, S. J. Frank, X. Zhang, X. R. Zhu, R. Mohan, Preliminary evaluation of multifield and single-field optimization for the treatment planning of spot-scanning proton therapy of head and neck cancer, Medical Physics 40 (8) (2013) 081709

  6. [5]

    Parodi, Vision 20/20: Positron emission tomography in radiation ther- apy planning, delivery, and monitoring, Medical Physics 42 (12) (2015) 7153–7168

    K. Parodi, Vision 20/20: Positron emission tomography in radiation ther- apy planning, delivery, and monitoring, Medical Physics 42 (12) (2015) 7153–7168

  7. [6]

    Paganetti, Proton therapy physics, Proton Therapy Physics

    H. Paganetti, Proton therapy physics, Proton Therapy Physics. Series: Se- ries in Medical Physics and Biomedical Engineering, ISBN: 978-1-4398- 3644-6. CRC Press, Edited by Harald Paganetti (12 2011). doi:10.1201/ b11448

  8. [7]

    Knopf, A

    A.-C. Knopf, A. Lomax, In vivo proton range verification: a review, Physics in Medicine & Biology 58 (15) (2013) R131

Show all 85 references
  1. [8]

    Paganetti, G

    H. Paganetti, G. El Fakhri, Monitoring proton therapy with PET, The British journal of radiology 88 (1051) (2015) 20150173

  2. [9]

    Espa˜ na, D

    S. Espa˜ na, D. S´ anchez-Parcerisa, P. Bragado, ´A. Guti´ errez-Uzquiza, A. Porras, C. Guti´ errez-Neira, A. Espinosa, V. V. Onecha, P. Ib´ a˜ nez, V. S´ anchez-Tembleque, et al., In vivo production of fluorine-18 in a chicken egg tumor model of breast cancer for proton thera...

  3. [10]

    K. Parodi, On-and off-line monitoring of ion beam treatment, Nuclear In- struments and Methods in Physics Research Section A: Accelerators, Spec- trometers, Detectors and Associated Equipment 809 (2016) 113–119. 23

  4. [11]

    Parodi, T

    K. Parodi, T. Yamaya, P. Moskal, Experience and new prospects of PET imaging for ion beam therapy monitoring, Zeitschrift f¨ ur Medizinische Physik 33 (1) (2023) 22–34

  5. [12]

    Bauer, D

    J. Bauer, D. Unholtz, F. Sommerer, C. Kurz, T. Haberer, K. Herfarth, T. Welzel, S. E. Combs, J. Debus, K. Parodi, Implementation and initial clinical experience of offline PET/CT-based verification of scanned carbon ion treatment, Radiotherapy and Oncology 107 (2) (2013) 218–226

  6. [13]

    Tomitani, J

    T. Tomitani, J. Pawelke, M. Kanazawa, K. Yoshikawa, K. Yoshida, M. Sato, A. Takami, M. Koga, Y. Futami, A. Kitagawa, et al., Washout studies of 11C in rabbit thigh muscle implanted by secondary beams of himac, Physics in Medicine & Biology 48 (7) (2003) 875

  7. [14]

    Mizuno, T

    H. Mizuno, T. Tomitani, M. Kanazawa, A. Kitagawa, J. Pawelke, Y. Iseki, E. Urakabe, M. Suda, A. Kawano, R. Iritani, et al., Washout measurement of radioisotope implanted by radioactive beams in the rabbit, Physics in Medicine & Biology 48 (15) (2003) 2269

  8. [16]

    Knopf, K

    A. Knopf, K. Parodi, T. Bortfeld, H. A. Shih, H. Paganetti, Systematic analysis of biological and physical limitations of proton beam range verifi- cation with offline PET/CT scans, Physics in Medicine & Biology 54 (14) (2009) 4477

  9. [17]

    Grogg, N

    K. Grogg, N. M. Alpert, X. Zhu, C. H. Min, M. Testa, B. Winey, M. D. Normandin, H. A. Shih, H. Paganetti, T. Bortfeld, et al., Mapping 15O production rate for proton therapy verification, International Journal of Radiation Oncology* Biology* Physics 92 (2) (2015) 453–459

  10. [18]

    Mart ´ ınez-Rovira, R

    I. Mart ´ ınez-Rovira, R. Boisgard, G. Pottier, B. Kuhnast, S. Jan, Experi- mental approach to evaluate the 11c perfusion and diffusion in small animal tissues for hadronPET applications, Plos one 11 (3) (2016) e0151212

  11. [19]

    Fiedler, M

    F. Fiedler, M. Priegnitz, R. J¨ ulich, J. Pawelke, P. Crespo, K. Parodi, F. P¨ onisch, W. Enghardt, In-beam PET measurements of biological half- lives of 12C irradiation induced β+-activity, Acta Oncologica 47 (6) (2008) 1077–1086

  12. [20]

    Nishio, A

    T. Nishio, A. Miyatake, T. Ogino, K. Nakagawa, N. Saijo, H. Esumi, The development and clinical use of a beam on-line PET system mounted on a rotating gantry port in proton therapy, International Journal of Radiation Oncology* Biology* Physics 76 (1) (2010) 277–286. 24

  13. [21]

    Toramatsu, A

    C. Toramatsu, A. Mohammadi, H. Wakizaka, N. Nitta, Y. Ikoma, C. Seki, I. Kanno, T. Yamaya, Tumour status prediction by means of carbon-ion beam irradiation: comparison of washout rates between in-beam PET and DCE-MRI in rats, Physics in Medicine & Biology 68 (19) (2023) 195005

  14. [22]

    M. R. Junttila, F. J. De Sauvage, Influence of tumour micro-environment heterogeneity on therapeutic response, Nature 501 (7467) (2013) 346–354

  15. [23]

    J. P. O’Connor, C. J. Rose, J. C. Waterton, R. A. Carano, G. J. Parker, A. Jackson, Imaging intratumor heterogeneity: role in therapy response, resistance, and clinical outcome, Clinical Cancer Research 21 (2) (2015) 249–257

  16. [24]

    Pugachev, S

    A. Pugachev, S. Ruan, S. Carlin, S. M. Larson, J. Campa, C. C. Ling, J. L. Humm, Dependence of FDG uptake on tumor microenvironment, In- ternational Journal of Radiation Oncology* Biology* Physics 62 (2) (2005) 545–553

  17. [25]

    Lovinfosse, M

    P. Lovinfosse, M. Hatt, D. Visvikis, R. Hustinx, Heterogeneity analysis of 18 F-FDG PET imaging in oncology: Clinical indications and perspectives, Clinical and Translational Imaging 6 (2018) 393–410

  18. [26]

    Dagogo-Jack, A

    I. Dagogo-Jack, A. T. Shaw, Tumour heterogeneity and resistance to cancer therapies, Nature reviews Clinical oncology 15 (2) (2018) 81–94

  19. [27]

    J. Wu, M. F. Gensheimer, X. Dong, D. L. Rubin, S. Napel, M. Diehn, B. W. Loo Jr, R. Li, Robust intratumor partitioning to identify high-risk subregions in lung cancer: a pilot study, International Journal of Radiation Oncology* Biology* Physics 95 (5) (2016) 1504–1512

  20. [28]

    Kim, S.-Y

    J. Kim, S.-Y. Ryu, S.-H. Lee, H. Y. Lee, H. Park, Clustering approach to identify intratumour heterogeneity combining FDG PET and diffusion- weighted MRI in lung adenocarcinoma, European radiology 29 (2019) 468– 475

  21. [29]

    H. Xu, W. Lv, H. Feng, D. Du, Q. Yuan, Q. Wang, Z. Dai, W. Yang, Q. Feng, J. Ma, et al., Subregional radiomics analysis of PET/CT imaging with intratumor partitioning: application to prognosis for nasopharyngeal carcinoma, Molecular Imaging and Biology 22 (2020) 1414–1426

  22. [30]

    Zhao, Y.-J

    X. Zhao, Y.-J. Liang, X. Zhang, D.-X. Wen, W. Fan, L.-Q. Tang, D. Dong, J. Tian, H.-Q. Mai, Deep learning signatures reveal multiscale intratumor heterogeneity associated with biological functions and survival in recur- rent nasopharyngeal carcinoma, European Journal of Nuclea...

  23. [31]

    Chitalia, V

    R. Chitalia, V. Viswanath, A. R. Pantel, L. M. Peterson, A. Gastounioti, E. A. Cohen, M. Muzi, J. Karp, D. A. Mankoff, D. Kontos, Functional 4-D clustering for characterizing intratumor heterogeneity in dynamic imag- ing: evaluation in FDG PET as a prognostic biomarker for bre...

  24. [32]

    M. Fan, P. Xia, R. Clarke, Y. Wang, L. Li, Radiogenomic signatures reveal multiscale intratumour heterogeneity associated with biological functions and survival in breast cancer, Nature communications 11 (1) (2020) 4861

  25. [33]

    Mansur, J

    A. Mansur, J. E. McConathy, E. Stringer-Reasor, G. Rocque, K. Khoury, N. Eltoum, M. Nikpanah, J. Bartels, B. Wright, N. Jahan, D. Jeffers, S. E. Lapi, A. G. Sorace, Quantitative [89zr]zr-trastuzumab pet and diffusion- weighted mri for characterization of metastatic her2-positi...

  26. [34]

    Cabrales, V

    P. Cabrales, V. V. Onecha, D. Izquierdo-Garc ´ ıa, L. M. Fraile, J. M. Ud ´ ıas, J. L. Herraiz, PROTOTWIN-PET: A deep learning and GPU- based workflow for dose verification in proton therapy with PET, IEEE Transactions on Radiation and Plasma Medical Sciences (2025) 1–1 doi: 1...

  27. [35]

    Izadi, D

    S. Izadi, D. Sutton, G. Hamarneh, Image denoising in the deep learning era, Artificial Intelligence Review 56 (7) (2023) 5929–5974

  28. [36]

    Valli` eres, E

    M. Valli` eres, E. Kay-Rivest, L. J. Perrin, X. Liem, C. Furstoss, N. Khaouam, P. F. Nguyen-Tan, C.-S. Wang, K. Sultanem, Data from Head-Neck-PET-CT, The Cancer Imaging Archive (2017). doi:10.7937/ K9/TCIA.2017.8oje5q00

  29. [37]

    S. M. Telloni, Tumor staging and grading: A primer, Molecular Profiling: Methods and Protocols (2017) 1–17

  30. [38]

    Ba la, Y

    K. Ba la, Y. Samovich, K. Dorobisz, Proton therapy in the treatment of head and neck cancers-review, Current Oncology Reports (2024) 1–8

  31. [39]

    Wieser, E

    H.-P. Wieser, E. Cisternas, N. Wahl, S. Ulrich, A. Stadler, H. Mescher, L.-R. M¨ uller, T. Klinge, H. Gabrys, L. Burigo, et al., Development of the open-source dose calculation and optimization toolkit matRad, Medical Physics 44 (6) (2017) 2556–2568

  32. [40]

    McNamara, A

    K. McNamara, A. Schiavi, D. Borys, K. Brzezinski, J. Gajewski, R. Kope´ c, A. Rucinski, T. Sk´ ora, S. Makkar, J. Hrbacek, et al., GPU accelerated Monte Carlo scoring of positron emitting isotopes produced during proton therapy for PET verification, Physics in Medicine & Biolo...

  33. [41]

    S. J. Frank, J. D. Cox, M. Gillin, R. Mohan, A. S. Garden, D. I. Rosenthal, G. B. Gunn, R. S. Weber, M. S. Kies, J. S. Lewin, et al., Multifield opti- mization intensity modulated proton therapy for head and neck tumors: a translation to practice, International Journal of Radi...

  34. [42]

    Suzuki, M

    K. Suzuki, M. B. Palmer, N. Sahoo, X. Zhang, F. Poenisch, D. S. Mackin, A. Y. Liu, R. Wu, X. R. Zhu, S. J. Frank, et al., Quantitative analysis of treatment process time and throughput capacity for spot scanning proton therapy, Medical Physics 43 (7) (2016) 3975–3986

  35. [43]

    URL https://www.nndc.bnl.gov/nudat3/

    National Nuclear Data Center, NuDat 3: Interactive Chart of Nuclides and Nuclear Structure and Decay Data, retrieved on: 2025-05-21 (2025). URL https://www.nndc.bnl.gov/nudat3/

  36. [44]

    Bauer, W

    J. Bauer, W. Chen, S. Nischwitz, J. Liebl, S. Rieken, T. Welzel, J. Debus, K. Parodi, Improving the modelling of irradiation-induced brain activation for in vivo PET verification of proton therapy, Radiotherapy and Oncology 128 (1) (2018) 101–108

  37. [45]

    Soppera, E

    N. Soppera, E. Dupont, M. Bossant, Janis book of proton induced cross sections, comparison of evaluated and experimental data from endf/b-vii. 1, Tech. rep., JENDL/HE-2007, PADF-2007, TENDL-2011 and EXFOR (2012)

  38. [46]

    Parodi, H

    K. Parodi, H. Paganetti, H. Shih, S. Michaud, J. Loeffler, T. Delaney, N. Liebsch, J. Munzenrider, A. Fischman, A. Knopf, T. Bortfeld, Patient study of in vivo verification of beam delivery and range, using positron emis- sion tomography and computed tomography imaging after p...

  39. [47]

    Toramatsu, A

    C. Toramatsu, A. Mohammadi, H. Wakizaka, C. Seki, F. Nishikido, S. Sato, I. Kanno, M. Takahashi, K. Karasawa, Y. Hirano, et al., Biological washout modelling for in-beam PET: rabbit brain irradiation by 11c and 15o ion beams, Physics in Medicine & Biology 65 (10) (2020) 105011

  40. [48]

    Meyer, Numpyrandomshapes3d: Generate 3d numpy array with random shapes, retrieved on: 2025-05-21 (2021)

    C. Meyer, Numpyrandomshapes3d: Generate 3d numpy array with random shapes, retrieved on: 2025-05-21 (2021). URL https://github.com/Cyril-Meyer/NumPyRandomShapes3D

  41. [49]

    S. Song, G. Zhang, Z. Yao, R. Chen, K. Liu, T. Zhang, G. Zeng, Z. Wang, R. Liu, Deep learning based on intratumoral heterogeneity predicts histopathologic grade of hepatocellular carcinoma, BMC Cancer 25 (1) (2025) 1–13

  42. [50]

    J. L. Herraiz, A. Lopez-Montes, A. Badal, MCGPU-PET: An open-source real-time Monte Carlo PET simulator, Computer Physics Communications 296 (2024) 109008

  43. [51]

    Schramm, K

    G. Schramm, K. Thielemans, PARALLELPROJ—an open-source frame- work for fast calculation of projections in tomography, Frontiers in Nuclear Medicine 3 (2024) 1324562

  44. [52]

    Gustafsson, J

    J. Gustafsson, J. Taprogge, On the reduction of imaging time-points for dosimetry in radionuclide therapy, EJNMMI physics 12 (1) (2025) 1–21. 27

  45. [53]

    Virtanen, R

    P. Virtanen, R. Gommers, T. E. Oliphant, M. Haberland, T. Reddy, D. Cournapeau, E. Burovski, P. Peterson, W. Weckesser, J. Bright, et al., SciPy 1.0: fundamental algorithms for scientific computing in Python, Na- ture methods 17 (3) (2020) 261–272

  46. [54]

    Cal-Gonz´ alez, S

    J. Cal-Gonz´ alez, S. Moore, M. Park, J. Herraiz, J. Vaquero, M. Desco, J. Udias, Improved quantification for local regions of interest in preclinical PET imaging, Physics in Medicine & Biology 60 (18) (2015) 7127

  47. [55]

    Lambert, F

    B. Lambert, F. Forbes, S. Doyle, H. Dehaene, M. Dojat, Trustworthy clini- cal AI solutions: a unified review of uncertainty quantification in deep learn- ing models for medical image analysis, Artificial Intelligence in Medicine (2024) 102830

  48. [56]

    Lemay, K

    A. Lemay, K. Hoebel, C. P. Bridge, B. Befano, S. De Sanjos´ e, D. Ege- men, A. C. Rodriguez, M. Schiffman, J. P. Campbell, J. Kalpathy-Cramer, Improving the repeatability of deep learning models with Monte Carlo dropout, npj Digital Medicine 5 (1) (2022) 174

  49. [57]

    Kendall, Y

    A. Kendall, Y. Gal, What uncertainties do we need in bayesian deep learn- ing for computer vision?, Advances in neural information processing sys- tems 30 (2017)

  50. [58]

    Seitzer, A

    M. Seitzer, A. Tavakoli, D. Antic, G. Martius, On the pitfalls of het- eroscedastic uncertainty estimation with probabilistic neural networks, arXiv preprint arXiv:2203.09168 (2022)

  51. [59]

    Z. Wang, A. C. Bovik, H. R. Sheikh, E. P. Simoncelli, Image quality as- sessment: from error visibility to structural similarity, IEEE transactions on image processing 13 (4) (2004) 600–612

  52. [60]

    H.-Y. Zhou, J. Guo, Y. Zhang, L. Yu, L. Wang, Y. Yu, nnformer: Interleaved transformer for volumetric segmentation, arXiv preprint arXiv:2109.03201 (2021)

  53. [61]

    Loshchilov, F

    I. Loshchilov, F. Hutter, Decoupled weight decay regularization, arXiv preprint arXiv:1711.05101 (2017)

  54. [62]

    Loshchilov, F

    I. Loshchilov, F. Hutter, Sgdr: Stochastic gradient descent with warm restarts, arXiv preprint arXiv:1608.03983 (2016)

  55. [63]

    Patro, K

    S. Patro, K. K. Sahu, Normalization: A preprocessing stage, arXiv preprint arXiv:1503.06462 (2015)

  56. [64]

    M. J. Cardoso, W. Li, R. Brown, N. Ma, E. Kerfoot, Y. Wang, B. Murrey, A. Myronenko, C. Zhao, D. Yang, et al., Monai: An open-source framework for deep learning in healthcare, arXiv preprint arXiv:2211.02701 (2022)

  57. [65]

    Craft, M

    D. Craft, M. Bangert, T. Long, D. Papp, J. Unkelbach, Shared data for intensity modulated radiation therapy (IMRT) optimization research: the CORT dataset, GigaScience 3 (1) (2014) 2047–217X. 28

  58. [66]

    Wilcoxon, Individual comparisons by ranking methods, in: Break- throughs in statistics: Methodology and distribution, Springer, 1992, pp

    F. Wilcoxon, Individual comparisons by ranking methods, in: Break- throughs in statistics: Methodology and distribution, Springer, 1992, pp. 196–202

  59. [67]

    Zhang, D

    X. Zhang, D. H. Pak, S. S. Ahn, X. Li, C. You, L. H. Staib, A. J. Sinusas, A. Wong, J. S. Duncan, Heteroscedastic uncertainty estimation framework for unsupervised registration, in: International Conference on Medical Im- age Computing and Computer-Assisted Intervention, Sprin...

  60. [68]

    Rusanov, G

    B. Rusanov, G. M. Hassan, M. Reynolds, M. Sabet, P. Rowshanfarzad, N. Bucknell, S. Gill, J. Dass, M. Ebert, Transformer CycleGAN with uncer- tainty estimation for CBCT based synthetic CT in adaptive radiotherapy, Physics in Medicine & Biology 69 (3) (2024) 035014

  61. [69]

    Schmitz, J

    J. Schmitz, J. Schwab, J. Schwenck, Q. Chen, L. Quintanilla-Martinez, M. Hahn, B. Wietek, N. Schwenzer, A. Staebler, U. Kohlhofer, et al., De- coding intratumoral heterogeneity of breast cancer by multiparametric in vivo imaging: a translational study, Cancer research 76 (18) ...

  62. [70]

    Schwenck, D

    J. Schwenck, D. Sonanini, J. M. Cotton, H.-G. Rammensee, C. la Foug` ere, L. Zender, B. J. Pichler, Advances in PET imaging of cancer, Nature Re- views Cancer 23 (7) (2023) 474–490

  63. [71]

    Salem, Hypoxia-targeted dose painting in radiotherapy, in: Seminars in Radiation Oncology, Vol

    A. Salem, Hypoxia-targeted dose painting in radiotherapy, in: Seminars in Radiation Oncology, Vol. 33, Elsevier, 2023, pp. 298–306

  64. [72]

    P. Vera, S. Thureau, F. Le Tinier, P. Chaumet-Riffaud, S. Hapdey, H. Kolesnikov-Gauthier, E. Martin, A. Berriolo-Riedinger, N. Pourel, J. M. Broglia, et al., Adaptive radiotherapy (up to 74 Gy) or standard radiother- apy (66 Gy) for patients with stage iii non-small-cell lung ...

  65. [73]

    Ramesh, D

    P. Ramesh, D. Ruan, S. J. Liu, Y. Seo, S. Braunstein, K. Sheng, Hypoxia- informed RBE-weighted beam orientation optimization for intensity mod- ulated proton therapy, Medical Physics 51 (3) (2024) 2320–2333

  66. [74]

    metabolism-guided

    G. Ferini, S. Parisi, S. Lillo, A. Viola, F. Minutoli, P. Critelli, V. Valenti, S. I. Illari, A. Brogna, G. E. Umana, et al., Impressive results after “metabolism-guided” lattice irradiation in patients submitted to palliative radiation therapy: Preliminary results of lattice ...

  67. [75]

    Ferini, V

    G. Ferini, V. Valenti, A. Tripoli, S. I. Illari, L. Molino, S. Parisi, A. Cac- ciola, S. Lillo, D. Giuffrida, S. Pergolizzi, Lattice or oxygen-guided radio- therapy: what if they converge? possible future directions in the era of immunotherapy, Cancers 13 (13) (2021) 3290. 29

  68. [76]

    Ten Eikelder, P

    S. Ten Eikelder, P. Ferjanˇ ciˇ c, A. Ajdari, T. Bortfeld, D. Den Hertog, R. Jeraj, Optimal treatment plan adaptation using mid-treatment imag- ing biomarkers, Physics in Medicine & Biology 65 (24) (2020) 245011

  69. [77]

    M. R. Horsman, L. S. Mortensen, J. B. Petersen, M. Busk, J. Overgaard, Imaging hypoxia to improve radiotherapy outcome, Nature reviews Clinical oncology 9 (12) (2012) 674–687

  70. [78]

    Boscolo, G

    D. Boscolo, G. Lovatti, O. Sokol, T. Vitacchio, F. Evangelista, E. Haettner, W. Tinganelli, C. Graeff, U. Weber, C. Schuy, et al., First image-guided treatment of a mouse tumor with radioactive ion beams, arXiv preprint arXiv:2409.14898 (2024)

  71. [79]

    Jouglar, L

    E. Jouglar, L. de Marzi, P. Verrelle, G. Cr´ ehange, R. Ferrand, F. Doz, Y. Prezado, X. Paoletti, From pre-clinical studies to human treatment with proton-minibeam radiation therapy: adapted idea, development, ex- ploration, assessment and long-term evaluation (ideal) framewor...

  72. [80]

    Geirnaert, L

    F. Geirnaert, L. Kerkhove, P. Montay-Gruel, T. Gevaert, I. Dufait, M. De Ridder, Exploring the metabolic impact of flash radiotherapy, Can- cers 17 (1) (2025) 133

  73. [81]

    Chung, I.-C

    Y.-H. Chung, I.-C. Cho, F. Jhang, C.-C. Weng, G. Lin, C.-F. Yu, F.-H. Chen, Dynamic profiles of early biological responses to predict the treat- ment efficacy of proton therapy in liver cancer assessed with in vivo ki- netic [18F]-FDG PET/MRI, Radiation Oncology 20 (1) (2025) ...

  74. [82]

    and Daube-Witherspoon, Mar- garet E

    Muller, Florence Marie and Li, Elizabeth J. and Daube-Witherspoon, Mar- garet E. and Pantel, Austin R. and Wiers, Corinde E. and Dubroff, Ja- cob G. and Vanhove, Christian and Vandenberghe, Stefaan and Karp, Joel S., Impact of deep learning denoising on kinetic modelling for l...

  75. [83]

    Cherezov, D

    D. Cherezov, D. Goldgof, L. Hall, R. Gillies, M. Schabath, H. M¨ uller, A. De- peursinge, Revealing tumor habitats from texture heterogeneity analysis for classification of lung cancer malignancy and aggressiveness, Scientific reports 9 (1) (2019) 4500

  76. [84]

    T. D. Malouff, A. Mahajan, S. Krishnan, C. Beltran, D. S. Seneviratne, D. M. Trifiletti, Carbon ion therapy: a modern review of an emerging technology, Frontiers in oncology 10 (2020) 82

  77. [85]

    Raschke, T

    F. Raschke, T. Wesemann, H. Wahl, S. Appold, M. Krause, J. Linn, E. Troost, Reduced diffusion in normal appearing white matter of glioma 30 patients following radio (chemo) therapy, Radiotherapy and Oncology 140 (2019) 110–115. 31

Pith tools

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