Pith. sign in

REVIEW 3 major objections 5 minor 50 references

COBRA2026: a large-scale multicenter pelvic cone-beam computed tomography projection dataset

T0 review · 3 major / 5 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read COBRA2026 is a 867-patient, six-center dataset of raw CBCT projections and matched CT-based references, created to give reconstruction and correction algorithms both clinical realism and aligned ground truth.

desk verdict Large, valuable measured-projection dataset, but the simulated-projection pairs have an undisclosed motion/reference mismatch. read the letter →

arxiv 2607.20037 v1 pith:NE3USE2Z submitted 2026-07-22 physics.med-ph

Adrian Thummerer , Simon Rit , Florian Kamp , Matteo Maspero , Martjin P.W. Intven , Thomas G. Boné , Christopher Kurz , Guillaume Landry
show 6 more authors
Thomas Baudier Mustafa Kadhim Julius Arnold Michael Rauter Barbara Knäusl Lukas Zimmermann
This is my paper · ORCID
classification physics.med-ph
keywords CBCTcone-beamcomputedtomographyradiotherapyprojectiondatadatasetreconstructiondeformableregistrationsyntheticCT
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

The paper presents COBRA2026, a publicly released dataset of raw cone-beam CT (CBCT) projection measurements from 867 patients who received pelvic radiotherapy at six European centers on Elekta and Varian systems. Each case includes the raw projections, acquisition geometry, calibration files, a clinically reconstructed CBCT, a planning CT deformably registered to the daily anatomy, and simulated projections computed from that deformed CT using the recorded geometry. The central aim is to provide researchers with a large, multicenter resource that combines the realism of clinical measurements with a geometrically consistent reference—something existing public radiotherapy CBCT datasets lack because they only provide reconstructed images. If the dataset is sound, it would let projection-domain deep-learning methods for tasks like sparse-view reconstruction, scatter correction, motion compensation, and synthetic CT generation be trained and evaluated at an unprecedented scale.

What carries the argument

The pre-processing pipeline is the central mechanism. It converts vendor-specific Elekta and Varian files into open formats; independently reconstructs a CBCT volume using FDK; deformably registers the planning CT to that volume using the IMPACT neural-network-based similarity metric plus an air-cavity matching step; and then simulates projections from the deformed CT using the recorded acquisition geometry, a randomized respiratory motion model, SpekPy-based spectrum modeling, scatter addition, and Poisson noise. The deformed CT with air-cavity correction is the load-bearing reference object that ties the measured and simulated projections together.

What would settle it

Pick a random subset of test cases, place fiducial markers or manually annotate corresponding anatomical landmarks in the deformed CT and the independently reconstructed CBCT, and measure the residual target registration error. If the median error exceeds roughly 3–5 mm (or a threshold appropriate for the task), the premise of a geometrically consistent reference collapses.

Watch

Extended reading notes

Core claim

The paper claims that COBRA2026 is a large-scale, multicenter collection of raw CBCT projection data plus paired simulated projections and deformably aligned planning CT images. It contains 867 clinical cases—692 training, 52 validation, and 123 test—totaling about 950 GB, acquired at six European centers using two vendor platforms. The key design choice is to release both measured projections (retaining clinical noise, scatter, motion, and system-specific artifacts) and simulated projections (generated from a deformably registered planning CT via a physics-based forward projector with matched angular sampling, bowtie filters, and Poisson noise). This pairing gives supervised-learning method

Load-bearing premise

The deformable registration and air-cavity matching produce a planning CT deformed closely enough to the day-of-treatment anatomy that it can serve as a reliable reference for simulated projections and for evaluating reconstruction quality.

Editorial extensions

If this is right

  • Researchers can train and benchmark projection-domain reconstruction and correction methods directly on raw clinical data, not just reconstructed images.
  • The six-center, two-vendor composition lets developers test whether algorithms generalize across institutions and scanner configurations.
  • The simulated projections provide paired image–projection ground truth for supervised learning, while measured projections preserve real artifacts that simulated data miss.
  • The dataset enables applications such as low-dose imaging, metal-artifact reduction, motion-compensated reconstruction, and synthetic CT generation from projection data.
  • Because the preprocessing pipeline is open source, other institutions can convert their own CBCT archives into the same format, expanding the dataset over time.

Reading between the lines

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

  • The validation and test splits were partly selected for high registration quality, so benchmark scores on them likely underestimate the difficulty of the most challenging clinical cases (those with large anatomical changes or severe artifacts).
  • The simulated projections are only as realistic as the forward model—effects like detector lag, afterglow, or a detailed detector response are not fully captured, so a domain gap between simulated and measured projections likely remains.
  • The deformed CT is not a true ground truth; residual registration errors and air-cavity approximations mean evaluation metrics computed against it carry an unknown bias. This argues for using the dataset mainly for relative comparisons between methods rather than absolute accuracy claims.
  • Because the dataset is pelvic-only, the authors' future extension to other anatomical sites—with different motion, scatter, and truncation characteristics—would be a natural follow-up that the pipeline already supports.
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 presents COBRA2026, described as the largest public radiotherapy CBCT dataset to date, comprising raw projection data, acquisition geometry, calibration information, clinically and independently reconstructed CBCTs, deformably registered planning CTs, and simulated projections for 867 pelvic patients from six European centers using Elekta and Varian systems. The preprocessing pipeline is open source and uses RTK, Elastix with a neural-network similarity metric, air-cavity matching, and the authors' simcbctgenerator for projection simulation. The dataset is split into training/validation/test sets and underpins the COBRA2026 reconstruction challenge. The authors claim the combination of measured and simulated projections provides geometrically aligned reference data for supervised learning and quantitative evaluation.

Significance. If the claims hold, this would be a valuable community resource: measured clinical CBCT projections at this scale are rare, and the release of open-format data, preprocessing scripts, and challenge infrastructure is a genuine strength. The paper is transparent about several limitations, including registration uncertainty, scatter-correction approximation, and selection bias. However, the central claim that the simulated projections form 'geometrically consistent image–projection pairs' with the released reference volumes is internally inconsistent because the simulation uses a randomized motion model while the reference is a single static deformed CT. In addition, the absence of quantitative validation of registration accuracy or simulation fidelity weakens the evidence that the deformed CT is a reliable geometric reference. The real-projection half of the dataset is still likely to be a useful resource, but the simulated half needs substantial revision or additional documentation before the paper's central claims can be fully supported.

major comments (3)
  1. [II.B.4, Table 3, Table 6] The simulated projections are generated with a 'randomized respiratory motion model' that displaces the abdominal wall, so each projection corresponds to a different phase of a moving anatomy. Yet the released reference for evaluation, ct_def.mha (Table 3) and ct.mha (Table 6), is a single static volume. No motion parameters, phase labels, or per-phase deformed CTs are included in the released files (Sections II.B.6, III.C). Consequently, a perfect reconstruction of the simulated projections cannot recover the static reference near the abdominal wall, and evaluation metrics on simulated data will penalize motion-induced mismatch rather than pure reconstruction error. This undermines the paper's claim that the simulated projections are 'geometrically consistent image–projection pairs' and is not disclosed in the Limitations (IV.C). Please either release the motion parameters and per-phase
  2. [II.C, IV.C] The dataset's value depends on (i) the deformed CT being an accurate geometric match to the CBCT anatomy and (ii) the simulated projections faithfully representing the clinical acquisition process. The paper relies exclusively on visual quality control and qualitative side-by-side comparisons; no quantitative registration errors (e.g., DICE or surface distance on bony anatomy, mean or percentile HU differences within the FOV) or projection-domain fidelity metrics (e.g., residuals between simulated and measured log projections) are reported. This lack of numbers makes it impossible for users to calibrate the reliability of the reference data and to judge whether residual errors are acceptable for the intended uses (reconstruction benchmarking, supervised learning). Please add quantitative validation, even on a subset, and report the inter-rater agreement of the visual QC if multiple revie
  3. [IV.C, II.B.4] The Limitations section discusses simplified physics and spatial-resolution limits of the simulated projections, but it does not acknowledge the specific and more serious issue that the randomized motion model makes the simulated projections inconsistent with the static reference volume that is released as ground truth. This is not merely a known limitation of the simulation; it is an internal inconsistency between two released file sets that directly affects how the challenge and any supervised evaluation on simulated data should be interpreted. The manuscript should either disclose this clearly in the Limitations and provide guidance on how users should handle the motion mismatch, or release the necessary motion information to make the simulated projections usable with the reference.
minor comments (5)
  1. [II.B.1] Typo: 'wihtout' should be 'without'.
  2. [II.B.2, Eq. (3)] Equation (3) defines CBCT_HU = CBCT_mu * 2^16 - 1024. Please clarify the definition of CBCT_mu and its calibration. As written, this does not look like the standard Hounsfield-unit conversion (which is linear in mu/mu_water). If this is a linear fit from Park et al., please state the calibration explicitly; otherwise, the equation may mislead users.
  3. [Figure 1 caption, Table 1] The text refers to 'centers A-D' and 'centers F and G' but there is no center E. Is this intentional (e.g., an excluded site) or a labeling error? Also, Table 1's 'Projection resolution [mm2]' unit is confusing; it seems to be pixel spacing ranges, not mm^2.
  4. [III.B, Table 5] The validation and test sets will not be released until March 2032. The abstract and dataset overview present the full 867-case split without emphasizing that only the training set is currently accessible. Please state clearly in the Data Format section what a user can immediately download and what remains embargoed.
  5. [II.C] The paper says cases with 'substantial deformable registration errors' were excluded, but no objective criterion or threshold is given. Please specify what 'substantial' means or at least describe the rating scale used in visual QC.

Circularity Check

2 steps flagged · score 4.0 of 10

Simulated projections are generated from the deformed CT that is also released as the challenge ground truth, making the synthetic evaluation self-referential; the randomized motion model further makes the static reference inconsistent with the moving simulated anatomy.

  1. self definitional [Section II.B.4 (Projection simulation) and Table 6 (Challenge ground truth)]
    ""To provide projection data paired with a known and geometrically consistent reference anatomy, cone-beam projections were simulated from the deformed CT using the acquisition geometry of the corresponding clinical scan." (II.B.4) "Ground Truth (used for evaluation on grand-challenge.org): ct.mha Deformed planning CT, masked to CBCT FOV" (Table 6)."

    The challenge ground truth for the simulated projections is the same deformed CT volume that was forward-projected to create those projections, so the evaluation target is the generation source by construction. A method that exactly inverts the authors' forward model would recover the reference, meaning the 'geometrically consistent reference' is definitional rather than an independent test of reconstruction accuracy. The randomized respiratory motion model in II.B.4 (abdominal-wall displacement per projection) additionally means the released static reference is not literally the source of the moving simulated projections, so the claimed consistency is both self-referential and inexact.

  2. self citation load bearing [Section II.B.4, Table 5, Reference 33]
    ""An acquisition was simulated for every gantry angle recorded in geometry.xmlusing the simcbctgenerator package (see Table 5) 33." Reference 33 is the same authors' preprint: Zimmermann et al., arXiv:2602.02130."

    The realism of the simulated projections, which underpin the claim of geometrically consistent reference data and the challenge's synthetic evaluation, is supported only by a citation to the authors' own preprint describing the same simulator. No external validation or independent benchmark of simcbctgenerator is provided in this paper, so the load-bearing assumption that the simulated projections adequately represent CBCT physics rests on self-citation. This is secondary to the self-definitional reference issue, and the simulator is open-source with a partially described physics model.

full rationale

COBRA2026 is primarily a data resource, and its core contribution—867 real clinical CBCT projection sets with acquisition geometry, calibration data, and reconstructed volumes—is not derived from a fitted model and remains externally anchored in the original vendor acquisitions. The measured-projection half of the dataset is therefore independent and useful. However, the simulated-projection half and the associated challenge evaluation contain a self-referential element: Section II.B.4 states that simulated projections were generated from the deformed CT via simcbctgenerator, and Table 6 lists the same deformed CT (ct.mha) as ground truth for evaluation. Thus the synthetic reconstruction target is, by construction, the volume used to generate the measurements; a perfect inversion of the authors' forward model would recover the reference, so the synthetic evaluation does not independently test reconstruction against an external truth. In addition, the randomized respiratory motion model in II.B.4 displaces the abdominal wall per projection, but the released reference is a single static volume; the paper's Limitations section (IV.C) acknowledges simplified physics and registration errors but does not disclose this motion/reference mismatch, which means the claimed 'geometrically consistent image–projection pairs' are not literally consistent. The self-citation to the authors' own simulator (ref 33) and the sCT model used in II.B.3 add to this concern, though the code is open-source and the physical model is partly specified. Because the central data resource (real projections) is independent and the circularity is confined to the synthetic evaluation chain, the score is 4 rather than higher.

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

The dataset depends on a chain of domain assumptions about vendor data handling, registration accuracy, and simulation realism. No new physical entities are postulated. The free parameters are calibration/processing choices, all disclosed in the text. The most consequential assumption is that the deformed CT is valid ground truth, since the simulated projections and evaluation reference derive from it.

free parameters (5)
  • Vendor gain factor g = Elekta: 0.851, Varian: 2.65
    Calibrated to match computed fluence with empirical reference exposures in the projection simulation (Section II.B.4). Affects simulated projection intensity and noise level.
  • Air-cavity threshold = -400 HU (Elekta), -200 HU (Varian)
    Chosen thresholds for air-cavity segmentation in the air-cavity matching procedure (Section II.B.3).
  • Motion model amplitude/frequency range = 2-7 mm, 12-20 cycles/min
    Randomized respiratory motion parameters used in simulated projections (Section II.B.4); chosen to approximate pelvic motion, not patient-specific.
  • Hann filter cut frequency = Elekta 0.99, Varian 0.4
    Chosen for the RTK FDK reconstruction of the CBCT reference (Section II.B.2).
  • Scatter correction parameters (Varian) = Not specified in detail
    Kernel-based scatter correction implementation following Sun et al., with parameters calibrated or approximated; the original implementation was not available (Section II.B.1, IV.C).
assumptions (5)
  • domain assumption Elekta exported projections have already undergone flood-field correction.
    Stated in Section II.B.1; relies on vendor export behavior and is not independently verified in the paper.
  • domain assumption Deformable registration using IMPACT with a pretrained TotalSegmentator feature layer aligns pCT to CBCT anatomy well enough to serve as a reference standard.
    Invoked in Section II.B.3; the paper reports no quantitative target registration error, only visual QC.
  • domain assumption Simulated projections using SpekPy spectra, a modelled bowtie profile, and Poisson noise adequately represent clinical CBCT acquisition for training and evaluation.
    Invoked in Section II.B.4; the authors explicitly state the simulation is a simplified approximation in Section IV.C.
  • domain assumption Visual quality control of overview figures is sufficient to identify and exclude cases with substantial registration errors.
    Invoked in Section II.C; this is subjective human review without inter-observer variability assessment.
  • domain assumption The kernel-based scatter correction for Varian acquisitions reduces shading artifacts without introducing significant bias.
    Invoked in Section II.B.1; the implementation is an independent approximation of Sun et al. and may differ from the original, as noted in Section IV.C.

how reviews work

0 comments
Cite this review

Pith. "Pith review of COBRA2026: a large-scale multicenter pelvic cone-beam computed tomography projection dataset." pith.science (2026). https://pith.science/paper/NE3USE2Z

@misc{pith2026260720037,
  author       = {Pith},
  title        = {Pith review of: COBRA2026: a large-scale multicenter pelvic cone-beam computed tomography projection dataset},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NE3USE2Z}},
  note         = {Machine review of arXiv:2607.20037}
}
read the original abstract

The COBRA2026 dataset is a large-scale, multicenter resource of raw radiotherapy cone-beam computed tomography (CBCT) acquisitions created for the development and evaluation of conventional and learning-based reconstruction and image-correction methods. It contains data from 867 patients undergoing pelvic radiotherapy at six European centers, acquired using Elekta and Varian imaging systems. For each case, the dataset includes raw projection data, acquisition geometry, calibration and correction information, clinically reconstructed CBCT images, and corresponding planning CT images. Vendor-specific files were anonymized and converted into open formats. Planning CT images were deformably registered to the daily CBCT anatomy, and matched projections were simulated using the corresponding acquisition geometry. All cases underwent visual quality control, and cases with substantial processing or registration errors were excluded. The approximately 950 GB dataset is divided into training, validation, and test sets containing 692, 52, and 123 cases, respectively. Projection stacks and volumetric images are provided as compressed MetaImage files, with geometry and metadata supplied in XML and YAML formats. COBRA2026 supports research on full- and sparse-view reconstruction, low-dose imaging, artifact and scatter correction, motion compensation, and synthetic CT generation. The dataset is released under the CC BY-NC 4.0 license, indexed on Zenodo (doi:10.5281/zenodo.21322350), and accompanied by openly available preprocessing and baseline reconstruction code. It also forms the basis of the COBRA2026 reconstruction challenge.

Figures

Figures reproduced from arXiv: 2607.20037 by the authors.

Figure 1
Figure 1. Representative axial pelvic CBCT images from the six participating centers, illustrating variability [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Overview of case C064 showing axial (first row), sagittal (second row), and coronal (third row) [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

50 extracted references · 2 canonical work pages

  1. [1]

    Journal of clinical oncology , volume=

    Advances in image-guided radiation therapy , author=. Journal of clinical oncology , volume=. 2007 , publisher=

    2007

  2. [2]

    2002 , publisher=

    Flat-panel cone-beam computed tomography for image-guided radiation therapy , author=. 2002 , publisher=

    2002

  3. [3]

    Physical and Engineering Sciences in Medicine , volume=

    A systematic review of volumetric image guidance in proton therapy , author=. Physical and Engineering Sciences in Medicine , volume=. 2023 , publisher=

    2023

  4. [4]

    2018 , publisher=

    Current state and future applications of radiological image guidance for particle therapy , author=. 2018 , publisher=

    2018

  5. [5]

    2019 , publisher=

    A review of cone-beam CT applications for adaptive radiotherapy of prostate cancer , author=. 2019 , publisher=

    2019

  6. [6]

    Journal of International Society of Preventive and Community Dentistry , volume=

    Artifacts: The downturn of CBCT image , author=. Journal of International Society of Preventive and Community Dentistry , volume=. 2015 , publisher=

    2015

  7. [7]

    2020 , publisher=

    A review of dose calculation approaches with cone beam CT in photon and proton therapy , author=. 2020 , publisher=

    2020

  8. [8]

    2021 , publisher=

    Deep learning based synthetic-CT generation in radiotherapy and PET: a review , author=. 2021 , publisher=

    2021

Show all 50 references
  1. [9]

    2023 , publisher=

    Segmentation by test-time optimization for CBCT-based adaptive radiation therapy , author=. 2023 , publisher=

    2023

  2. [10]

    2017 , publisher=

    Optimal combination of anti-scatter grids and software correction for CBCT imaging , author=. 2017 , publisher=

    2017

  3. [11]

    2024 , publisher=

    Image quality evaluation of a new high-performance ring-gantry cone-beam computed tomography imager , author=. 2024 , publisher=

    2024

  4. [12]

    2016 , publisher=

    Motion-aware temporal regularization for improved 4D cone-beam computed tomography , author=. 2016 , publisher=

    2016

  5. [13]

    2009 , publisher=

    Iterative image reconstruction for CBCT using edge-preserving prior , author=. 2009 , publisher=

    2009

  6. [14]

    2020 , publisher=

    Comparison of CBCT based synthetic CT methods suitable for proton dose calculations in adaptive proton therapy , author=. 2020 , publisher=

    2020

  7. [15]

    IEEE Transactions on Radiation and Plasma Medical Sciences , volume=

    Synthetic CT image generation from CBCT: a systematic review , author=. IEEE Transactions on Radiation and Plasma Medical Sciences , volume=. 2025 , publisher=

    2025

  8. [16]

    2019 , publisher=

    Projection-domain scatter correction for cone beam computed tomography using a residual convolutional neural network , author=. 2019 , publisher=

    2019

  9. [17]

    2019 , publisher=

    SPARE: Sparse-view reconstruction challenge for 4D cone-beam CT from a 1-min scan , author=. 2019 , publisher=

    2019

  10. [18]

    2023 , publisher=

    SynthRAD2023 Grand Challenge dataset: Generating synthetic CT for radiotherapy , author=. 2023 , publisher=

    2023

  11. [19]

    2025 , publisher=

    SynthRAD2025 Grand Challenge dataset: Generating synthetic CTs for radiotherapy from head to abdomen , author=. 2025 , publisher=

    2025

  12. [20]

    Journal of Physics: Conference Series , volume=

    The Reconstruction Toolkit (RTK), an open-source cone-beam CT reconstruction toolkit based on the Insight Toolkit (ITK) , author=. Journal of Physics: Conference Series , volume=. 2014 , doi=

    2014

  13. [21]

    Proceedings of the XXth International Conference on the use of Computers in Radiation therapy (ICCR) , year=

    Reconstruction Toolkit (RTK) v2, an Insight Toolkit (ITK) module for tomographic reconstruction , author=. Proceedings of the XXth International Conference on the use of Computers in Radiation therapy (ICCR) , year=

  14. [22]

    Journal of the Optical Society of America A , volume=

    Practical cone-beam algorithm , author=. Journal of the Optical Society of America A , volume=. 1984 , publisher=

    1984

  15. [23]

    2010 , doi=

    Improved scatter correction using adaptive scatter kernel superposition , author=. 2010 , doi=

    2010

  16. [24]

    1982 , publisher=

    Optimal short scan convolution reconstruction for fan beam CT , author=. 1982 , publisher=

    1982

  17. [25]

    2002 , publisher=

    X-ray micro-CT with a displaced detector array , author=. 2002 , publisher=

    2002

  18. [26]

    Proceedings of the IEEE , volume =

    On the Use of Windows for Harmonic Analysis with the Discrete Fourier Transform , author =. Proceedings of the IEEE , volume =. 1978 , doi =

    1978

  19. [27]

    Proceedings of the IEEE conference on computer vision and pattern recognition workshops , pages=

    SimpleElastix: A user-friendly, multi-lingual library for medical image registration , author=. Proceedings of the IEEE conference on computer vision and pattern recognition workshops , pages=. 2016 , doi=

    2016

  20. [28]

    IEEE transactions on medical imaging , volume=

    Elastix: a toolbox for intensity-based medical image registration , author=. IEEE transactions on medical imaging , volume=. 2009 , publisher=

    2009

  21. [29]

    arXiv preprint arXiv:2503.24121 , year=

    IMPACT: a generic semantic loss for multimodal medical image registration , author=. arXiv preprint arXiv:2503.24121 , year=

    arXiv

  22. [30]

    arXiv preprint arXiv:2602.02130 , year=

    Eliminating Registration Bias in Synthetic CT Generation: A Physics-Based Simulation Framework , author=. arXiv preprint arXiv:2602.02130 , year=

  23. [31]

    2024 , publisher=

    Minimum imaging dose for deep learning-based pelvic synthetic computed tomography generation from cone beam images , author=. 2024 , publisher=

    2024

  24. [32]

    2021 , publisher=

    Reducing axial truncation artifacts in iterative cone-beam CT for radiation therapy using a priori preconditioned information , author=. 2021 , publisher=

    2021

  25. [33]

    European Journal of Radiology , volume=

    Metal artifact reduction using iterative CBCT reconstruction algorithm for head and neck radiation therapy: a phantom and clinical study , author=. European Journal of Radiology , volume=. 2020 , publisher=

    2020

  26. [34]

    2009 , publisher=

    On-the-fly motion-compensated cone-beam CT using an a priori model of the respiratory motion , author=. 2009 , publisher=

    2009

  27. [35]

    2025 , publisher=

    Deep learning based super-resolution for CBCT dose reduction in radiotherapy , author=. 2025 , publisher=

    2025

  28. [36]

    doi:10.5281/zenodo.21397775 , url =

    Thummerer, Adrian and Zimmermann, Lukas and COBRA2026-contributors , title =. doi:10.5281/zenodo.21397775 , url =

    doi

  29. [37]

    doi:10.5281/zenodo.21397870 , url =

    Thummerer, Adrian and Zimmermann, Lukas and COBRA2026-contributors , title =. doi:10.5281/zenodo.21397870 , url =

    doi

  30. [38]

    Seminars in radiation oncology , volume=

    Adaptive radiotherapy for anatomical changes , author=. Seminars in radiation oncology , volume=. 2019 , organization=

    2019

  31. [39]

    2017 , publisher=

    Online adaptive radiation therapy , author=. 2017 , publisher=

    2017

  32. [40]

    2022 , publisher=

    Deep learning methods for enhancing cone-beam CT image quality toward adaptive radiation therapy: a systematic review , author=. 2022 , publisher=

    2022

  33. [41]

    Journal of applied clinical medical physics , volume=

    Assessment of CBCT--based synthetic CT generation accuracy for adaptive radiotherapy planning , author=. Journal of applied clinical medical physics , volume=. 2022 , publisher=

    2022

  34. [42]

    2016 , publisher=

    Investigating deformable image registration and scatter correction for CBCT-based dose calculation in adaptive IMPT , author=. 2016 , publisher=

    2016

  35. [43]

    2017 , publisher=

    A longitudinal four-dimensional computed tomography and cone beam computed tomography dataset for image-guided radiation therapy research in lung cancer , author=. 2017 , publisher=

    2017

  36. [44]

    0—a software toolkit for modeling x-ray tube spectra , author=

    SpekPy v2. 0—a software toolkit for modeling x-ray tube spectra , author=. 2021 , publisher=

    2021

  37. [45]

    Frontiers in neuroinformatics , volume=

    ITK: enabling reproducible research and open science , author=. Frontiers in neuroinformatics , volume=. 2014 , publisher=

    2014

  38. [46]

    Scientific Data , volume=

    CT and cone-beam CT of ablative radiation therapy for pancreatic cancer with expert organ-at-risk contours , author=. Scientific Data , volume=. 2022 , publisher=

    2022

  39. [47]

    Scientific data , volume=

    A cone-beam X-ray computed tomography data collection designed for machine learning , author=. Scientific data , volume=. 2019 , publisher=

    2019

  40. [48]

    Medical Image Analysis , pages=

    Multi-structure segmentation in CBCT volumes: The ToothFairy2 challenge , author=. Medical Image Analysis , pages=. 2026 , publisher=

    2026

  41. [49]

    2024 IEEE International Conference on Acoustics, Speech, and Signal Processing Workshops (ICASSPW) , pages=

    Advancing the frontiers of deep learning for low-dose 3d cone-beam computed tomography (ct) reconstruction , author=. 2024 IEEE International Conference on Acoustics, Speech, and Signal Processing Workshops (ICASSPW) , pages=. 2024 , organization=

    2024

  42. [50]

    Medical physics , volume=

    Proton dose calculation on scatter-corrected CBCT image: feasibility study for adaptive proton therapy , author=. Medical physics , volume=. 2015 , publisher=

    2015

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.