Pith. sign in

REVIEW 3 major objections 3 minor 19 references

Cross-patient transfer of anatomical change predicts treatment-day CT better than the planning scan.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-05 00:11 UTC pith:BC6ON7UG

load-bearing objection A genuinely novel two-step cross-patient motion transfer, undermined by an oracle prior-selection step that makes the headline numbers upper bounds; worth a serious referee, not acceptance as is. the 3 major comments →

arxiv 2608.00831 v1 pith:BC6ON7UG submitted 2026-08-01 physics.med-ph cs.AI

Anticipatory Digital Twins for Online Head-and-Neck Adaptive Proton Therapy via Foundation-Model Registration

Yizhou Wu , Yuheng Li , Xiaofeng Yang , Chih-Wei Chang This is my paper
classification physics.med-ph cs.AI
keywords digital twinadaptive proton therapyhead and neck cancerdeformable image registrationfoundation modelpredicted CTcross-patient motion transferonline adaptive radiotherapy
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper tries to establish that a head-and-neck proton therapy patient's anatomy on a treatment day can be anticipated before any new scan is taken, by transferring the longitudinal anatomical change of a previously treated patient onto the target patient's own planning CT. The proposed digital-twin pipeline performs two deformable registrations with a frozen pretrained foundation model: one aligns the prior patient to the target, the other re-expresses the prior's planning-to-QACT change in the target's frame, producing predicted CTs with propagated organ contours. Tested on 88 patients (20 held-out targets, each with three QACTs), the predicted CTs beat the static planning CT on every agreement metric—NCC up 22.8%, OAR Dice up 20.2%, and CT-number error down 23.4%—with the largest gains in patients whose anatomy changed most. If the claim holds, adaptive proton therapy could shift from reactive multi-day replanning to proactive selection from a precomputed plan library. The authors note that the primary results rank prior patients by an oracle score against the held-out QACT, so the deployment-time prior selector remains the unproven step.

Core claim

Central claim: anatomical change during head-and-neck radiotherapy is canonical enough across patients (tumor regression, weight loss, parotid migration) that a prior patient's planning-to-treatment-day change, re-expressed on a new target by one cross-patient registration, predicts the target's own treatment-day anatomy better than the target's static planning CT. On the hardest case—the QA CT with the largest change—across 20 targets, predicted CTs improved normalized cross-correlation by 22.8%, OAR Dice by 20.2%, cut mean CT-number error by 23.4%, and raised a six-metric composite from 0.77 to 0.82. An ablation shows that transferring the prior's anatomy directly degrades structural metri

What carries the argument

The carrying mechanism is a two-step cross-patient longitudinal motion transfer producing two deformation vector fields. DVF1 (φ1 = R(It, Ip)) aligns the prior patient's planning CT to the target planning CT, carrying the prior's three QA CTs into the target frame; DVF2 (φ2(k) = R(Q̃p(k), Ĩp)) estimates the prior's planning-to-QA change in that frame; applying φ2(k) to the target's own planning CT synthesizes three predicted CTs with propagated contours. This keeps the target's anatomy while importing another patient's longitudinal change. All registrations use one frozen pretrained foundation model (FlexiCT-2D), applied slice-wise to 2D CT patches and reused without patient-specific trainin

Load-bearing premise

The paper's main results pick a prior patient using the held-out future scan as an oracle; the load-bearing premise is that a purely image-similarity score, available before that scan exists, will pick a prior whose transferred change still beats the static planning CT—and this premise is not yet tested.

What would settle it

Run the full pipeline with prior patients chosen by the composite similarity score instead of the oracle, on a held-out cohort, and count how often the resulting pdCT beats the planning CT on NCC, MAE, and OAR Dice against the actual QACT; if the median per-patient gains drop to zero or negative, the claimed anticipatory advantage is not realizable in practice. A complementary dose-domain falsifier: compute robust proton plans on the three pdCTs and compare dose-volume histogram metrics with the plan actually delivered on the QACT; the prediction is only clinically useful if dose on predicted

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • pdCTs and contours exist before any treatment-day image is acquired, so a precomputed plan library can be selected on the day instead of waiting days for offline replanning.
  • Patients with the largest anatomical change—where no adaptation fails worst—gain the most, while patients with stable anatomy need not be re-planned at all.
  • Because the registration model is frozen and pretrained, the method needs no patient-specific training and runs in roughly 22 minutes per prediction on a GPU, making online use plausible.
  • Multiple pdCTs per prior-target pairing (one per QACT) supply a natural uncertainty set for robust treatment planning.
  • The composite quality metric provides a single score for comparing prospective predicted CTs, enabling quality gating of synthesized anatomy before clinical use.

Where Pith is reading between the lines

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

  • An untested corollary of the canonical-change hypothesis: the same two-step transfer could predict treatment-day anatomy from planning CT alone in other sites with stereotyped change (prostate, lung), given a prior longitudinal scan library—the authors only demonstrate head-and-neck.
  • A concrete next experiment the paper leaves open: replace the oracle prior ranking with the composite similarity score, then measure how often the automatically selected prior's pdCT still beats the planning CT; this would quantify the true deployable gain.
  • The three pdCTs per target can be read as an uncertainty set; a robust proton plan optimized over them would provide a dose-domain check of whether predicted anatomy is good enough for treatment decisions.
  • Because the ablation shows borrowed anatomy hurts, the method's ceiling depends on matching the direction of change, not just image similarity; a learned selector trained on retrospective transfer outcomes could push performance beyond the oracle.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 3 minor

Summary. The paper proposes a digital-twin framework for head-and-neck proton therapy that synthesizes predicted treatment-day CTs (pdCTs) before any treatment-day image is acquired. For a target patient, a prior patient's planning CT is registered to the target (DVF1) and the prior's planning-to-QACT deformation is re-estimated on the target grid (DVF2) and applied to the target's own planning CT, yielding three pdCTs with propagated contours. The registration model is a frozen foundation model (FlexiCT-2D) used without patient-specific training. On 88 HN patients (20 targets), the reported pdCTs improve over the static planning CT in NCC (+22.8%), Dice (+20.2%), MAE (-23.4%), HD95, ASSD, and a composite score. The authors also report an ablation showing that transferring the prior patient's change, rather than the prior's anatomy, is essential. The central claim is that cross-patient longitudinal change transfer can anticipate treatment-day anatomy before imaging.

Significance. If the anticipatory claim were actually demonstrated, this would be a valuable contribution to online adaptive proton therapy: it would enable precomputed plan libraries and shift adaptation from reactive to proactive, potentially reducing the need for repeated imaging. The paper has several strengths: it uses a fully frozen foundation-model registration pipeline, avoids per-patient training, evaluates on a non-trivial cohort of 88 patients with 5,220 DVF2 pairings, and includes an ablation that isolates the contribution of the longitudinal-change step. However, the reported headline gains are obtained with an oracle prior-patient selector that ranks priors against the held-out QACT, and the evaluation further selects the largest-change QACT per target using ground truth. These choices mean the current experiments do not support the paper's central 'before image acquisition' claim; they establish an upper bound on what a deployable selector might achieve. The paper's own Discussion acknowledges this gap and defers a ground-truth-free selector to future work, which is candid but does not resolve the mismatch between the abstract's claim and the evidence.

major comments (3)
  1. [Section 3.3 and Table 1] The primary quantitative results use an oracle prior-patient selector: candidate priors are ranked against the held-out QACT, which is unavailable at prediction time. The composite NCC/LPIPS/MI score described in Sec. 2.1 is never used in the reported comparisons, and there is no evidence that a ground-truth-free selector would recover the gains in Table 1. The abstract and conclusions therefore overstate the 'anticipatory' capability. The authors should either (a) evaluate the top-K composite-selected prior and report the resulting metrics, or (b) explicitly reframe the results as an upper bound. Without this, the central claim that treatment-day anatomy can be predicted before imaging is unsupported.
  2. [Section 2.4 / Section 3.1] The evaluation selects, for each target, the QACT with the largest change from the planning CT. This is a ground-truth-based choice of the evaluation time point. It cherry-picks the most favorable scenario (and the one where gains are largest, as the authors state) and does not represent the full distribution of treatment days. Please report results for all three QACTs separately or aggregated, or provide a clear clinical rationale for why the largest-change QACT is the appropriate single evaluation point. As written, the reported improvements may reflect the selection of the time point rather than the predictive quality of the method.
  3. [Section 2.1] The proposed deployable prior-patient selector is a composite of NCC, LPIPS, and MI computed between the target's planning CT and each prior's planning CT. This selector is never quantitatively evaluated. It is unclear whether static planning-CT similarity identifies priors whose longitudinal change transfers usefully to the target. The paper needs at least a correlation or ablation showing how often the composite-selected prior matches the oracle-selected prior, and the downstream pdCT metrics for the composite-selected prior. Without this, the method has no demonstrated deployment path.
minor comments (3)
  1. [Section 2.2] There is a typo: 'The output of λ (λ = 1). The output of R is φ in voxel units...' The first phrase is a duplicated fragment. Also, the top-K value K in Sec. 2.1 is never specified; please state how many priors are retained and how the 'primary' prior is distinguished from the rest.
  2. [Section 2.4] The dataset source is 'masked for anonymization' and the ethics statement is deferred. For reproducibility and journal requirements, please provide the data availability and ethics approval details, or a clear explanation of the anonymization constraints.
  3. [Section 3.1 / Table 1] The 'Improved' column is defined as per-patient relative gain, but the table only shows mean values. Please include the standard deviation or confidence interval for the per-patient gains, or clarify how the percentage was computed.

Circularity Check

2 steps flagged

Reported pdCT gains rely on oracle prior-patient selection against the held-out QACT; the deployable composite selector is never evaluated, so the central anticipatory claim is a test-set-fitted upper bound.

specific steps
  1. fitted input called prediction [Section 3.3 (Discussion); Table 1; Section 2.1]
    "Our primary results rank candidate priors against the held-out QACT, an oracle unavailable at instantiation time. The composite image-similarity score (NCC, LPIPS, MI) is a step toward a deployable selector, but closing the gap to the oracle ranking will likely require a learned, ground-truth-free scorer that predicts which priors carry transferable change for a given target."

    The pdCTs in Table 1 are generated from the prior patient selected by ranking against the very QACT on which the pdCT is then evaluated. Thus the reported 22.8% NCC, 20.2% Dice, and 23.4% MAE gains are not predictions of an unseen treatment-day anatomy; they are results of a selection rule that already used that anatomy. The proposed ground-truth-free composite selector is described but never used in the primary comparison, so the paper's central claim of anticipating treatment-day anatomy before image acquisition is not demonstrated independently of the target outcome.

  2. other [Section 3.1 (Quantitative agreement)]
    "For each target we evaluate the most challenging treatment-day anatomy, the QACT with the largest change from the planning CT, where no adaptation fails worst and prediction matters most."

    The evaluation point is also chosen using the ground-truth QACTs: per target, the QACT with the largest planning-to-QACT change is selected for reporting. This uses held-out treatment-day data to pick the timepoint most favorable to the method, compounding the oracle-selection problem. It is not a direct definitional reduction, but it further biases the reported improvement toward the setting where the static baseline is weakest.

full rationale

The two-step registration equations (1)-(3) are not themselves circular: the predicted CT is φ2∘It, and φ2 is estimated from a prior patient's aligned planning/QACT pair, independent of the target QACT. The circularity is in the evaluation protocol. Because the prior-patient choice in Table 1 is an oracle ranking against the held-out QACT, the comparison 'pdCT vs QACT' is partly self-fulfilling. The paper honestly discloses this in Section 3.3, but the abstract and conclusions claim an anticipatory capability that the experiments do not establish. No load-bearing self-citation circularity was found: FlexiCT-2D [20] is used as a frozen external component, and the self-citations to prior digital-twin work are contextual rather than the source of the quantitative claim. The score reflects that the central evidence is oracle-fitted, not that the generative equations are definitionally circular.

Axiom & Free-Parameter Ledger

2 free parameters · 5 axioms · 0 invented entities

The central claim rests on the assumed canonicality of anatomical change, the accuracy of a frozen foundation registration model, and, critically, the existence of a deployable prior selector that the paper does not validate. The oracle selection used in the experiments is not an axiom but a test-set leak, captured in the red flags.

free parameters (2)
  • lambda (diffusion regularization weight) = 1
    Weight on the first-order diffusion penalty in the registration objective (Section 2.2); fixed by hand across all experiments, not tuned to data.
  • top-K prior selection (K)
    The number of retained priors is not specified; only the most similar prior is used for the primary prediction. This discrete parameter is undefined.
axioms (5)
  • domain assumption Anatomical change during head-and-neck radiotherapy is largely canonical across patients (tumor regression, weight loss, medial parotid migration)
    Stated in the Discussion ('this succeeds because anatomical change... is largely canonical across patients'); if false, cross-patient transfer would be uninformative.
  • domain assumption FlexiCT-2D foundation model, pretrained on public CT volumes and applied frozen, produces accurate deformable registration for both cross-patient (φ1) and longitudinal (φ2) alignments without patient-specific training
    The entire pipeline relies on this model's outputs (Section 2.2); no validation of registration accuracy on the institutional cohort is provided.
  • domain assumption The three QACTs per patient are valid ground truth for treatment-day anatomy after rigid alignment to the planning grid
    Section 2.4 treats QACTs as treatment-day ground truth; acquisition geometry and timing are not detailed.
  • domain assumption The two-step composition is physically plausible: φ1 removes inter-patient differences so φ2(k) captures longitudinal change rather than identity, and applying φ2(k) to It yields a valid anatomy
    Section 2.3 states this design rationale; only a brief mention of low folding fraction without reported numbers.
  • ad hoc to paper A ground-truth-free prior selector can replace the oracle ranking without losing the reported gains
    The Discussion states this is the 'dominant remaining bottleneck' and future work; the primary results do not use such a selector.

pith-pipeline@v1.3.0-alltime-deepseek · 7361 in / 15891 out tokens · 142972 ms · 2026-08-05T00:11:11.100850+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of Anticipatory Digital Twins for Online Head-and-Neck Adaptive Proton Therapy via Foundation-Model Registration." pith.science (2026). https://pith.science/paper/BC6ON7UG

@misc{pith2026260800831,
  author       = {Pith},
  title        = {Pith review of: Anticipatory Digital Twins for Online Head-and-Neck Adaptive Proton Therapy via Foundation-Model Registration},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BC6ON7UG}},
  note         = {Machine review of arXiv:2608.00831}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Head-and-neck (HN) proton therapy is highly sensitive to anatomical change over a 4-to-6-week course, as tumor shrinkage, weight loss, and setup variation can misposition the Bragg peak near critical organs such as the parotids, oral cavity, brainstem, and spinal cord, leading to target underdosing or organ-at-risk overdosing. Online adaptive proton therapy replans on the anatomy of the day, yet standard workflows rely on offline replanning that requires repeated CT acquisition and roughly a week of preparation, adding burden, cost, and delay. We investigate whether a patient's treatment-day anatomy can be predicted before image acquisition by transferring longitudinal change from a population database. We propose a digital-twin framework built on a pretrained foundation-model deformable registration network used without patient-specific training. A first registration aligns a prior patient's planning CT to the target and carries the prior's during-treatment quality assurance CT (QACT) into the target frame; a second registration estimates the prior's planning-to-QACT change, which is then applied to the target's own planning CT to synthesize predicted CTs (pdCTs) with propagated contours. Using 88 HN patients, each with a planning CT and three QACTs, we show that pdCTs better match treatment-day anatomy than the static planning CT. Compared with the planning CT alone, normalized cross-correlation improves by 22.8%, Dice for organs-at-risk by 20.2%, and CT-number error decreases by 23.4%. Gains are largest for patients with major anatomical change and negligible when anatomy is stable. This cross-patient motion transfer leverages the digital-twin concept to anticipate treatment-day anatomy, enabling personalized online adaptive proton therapy without repeated imaging.

Figures

Figures reproduced from arXiv: 2608.00831 by Chih-Wei Chang, Xiaofeng Yang, Yizhou Wu, Yuheng Li.

Figure 1
Figure 1. Figure 1: Proposed digital-twin framework for predicted-CT generation. A prior patient's planning CT is aligned to the target (φ₁, DVF₁), bringing the prior QACTs into the target frame; the prior's planning-to-QACT change is estimated per QACT (φ₂(k), DVF₂) and applied to the target plan￾ning CT to synthesize three pdCTs, with contours propagated by the same fields. 2.1 Prior patient selection and atlas ranking All … view at source ↗
Figure 2
Figure 2. Figure 2: CT quality and error for a representative target patient, on a coronal slice. Col￾umns: planning CT (TPCT), predicted CT (pdCT), treatment-day QACT, and absolute￾TPCT pCT QACT |TPCT – QACT| |pCT – QACT| [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Propagated OAR contours on the treatment-day QACT for a representative target. Rows: TPCT (no adaptation) vs pdCT; columns: brainstem, spinal cord, pharyngeal constrictor. Or￾ange/green = overlap with the QACT for TPCT / pdCT; blue outline = QACT; per-panel Dice annotated. 3.2 Ablations Removing the longitudinal-change step. We compare the full method (pdCT) against an ablated variant, pdCT no φ₂, that ski… view at source ↗
Figure 4
Figure 4. Figure 4: Ablation: full method (pdCT) vs. the variant without the longitudinal-change step (pdCT no φ₂), mean ± SD over 20 patients. Using the φ₁-aligned prior QACT di￾rectly raises NCC but degrades all structural metrics, while pdCT is best on every met￾ric [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

19 extracted references · 14 canonical work pages

  1. [1]

    Anatomic changes in head and neck intensity-modulated proton therapy: Comparison between robust optimization and online adaptation

    Lalonde A, Bobić M, Winey B, Verburg J, Sharp GC, Paganetti H. Anatomic changes in head and neck intensity-modulated proton therapy: Comparison between robust optimization and online adaptation. Radiother Oncol. 2021;159:39–47. doi: 10.1016/j.radonc.2021.03.008

    work page doi:10.1016/j.radonc.2021.03.008 2021

  2. [2]

    A retrospective study on the investigation of potential dosimetric benefits of online adaptive proton therapy for head and neck cancer

    Chang CW, Bohannon D, Tian Z, Wang Y, McDonald MW, Yu DS, et al. A retrospective study on the investigation of potential dosimetric benefits of online adaptive proton therapy for head and neck cancer. J Appl Clin Med Phys. 2024;25(5):e14308. doi: 10.1002/acm2.14308

    work page doi:10.1002/acm2.14308 2024

  3. [3]

    Adaptive proton therapy

    Paganetti H, Botas P, Sharp GC, Winey B. Adaptive proton therapy. Phys Med Biol. 2021;66(22). doi: 10.1088/1361-6560/ac344f

    work page doi:10.1088/1361-6560/ac344f 2021

  4. [4]

    Digital twins for health: a scoping review

    Katsoulakis E, Wang Q, Wu H, Shahriyari L, Fletcher R, Liu J, et al. Digital twins for health: a scoping review. npj Digital Medicine. 2024;7(1):77. doi: 10.1038/s41746-024-01073-0

    doi:10.1038/s41746-024-01073-0 2024

  5. [5]

    Exploration of an adaptive proton therapy strategy using CBCT with the concept of digital twins

    Chang CW, Tian Z, Qiu RLJ, Scott Mcginnis H, Bohannon D, Patel P, et al. Exploration of an adaptive proton therapy strategy using CBCT with the concept of digital twins. Phys Med Biol. 2025;70(2). doi: 10.1088/1361-6560/ada684

    work page doi:10.1088/1361-6560/ada684 2025

  6. [6]

    The ob-jective is a sum-of-squared-differences feature loss plus a first-order diffusion penalty weighted by λ (λ = 1)

    on the features (downsampled 2×), with H, W, D the image height, width, and depth and the last dimension the three displacement components. The ob-jective is a sum-of-squared-differences feature loss plus a first-order diffusion penalty weighted by λ (λ = 1). The output of λ (λ = 1). The output of R is φ in voxel units; image and label warps use trilinear...

    work page 2000

  7. [7]

    Digital twins for the era of personalized surgery

    Mekki YM, Luijten G, Hagert E, Belkhair S, Varghese C, Qadir J, et al. Digital twins for the era of personalized surgery. npj Digital Medicine. 2025;8(1):283. doi: 10.1038/s41746-025-01575-5

    doi:10.1038/s41746-025-01575-5 2025

  8. [8]

    Digital twins, synthetic patient data, and in-silico trials: can they empower paediatric clinical trials? The Lancet Digital Health

    Pammi M, Shah PS, Yang LK, Hagan J, Aghaeepour N, Neu J. Digital twins, synthetic patient data, and in-silico trials: can they empower paediatric clinical trials? The Lancet Digital Health. 2025;7(5). doi: 10.1016/j.landig.2025.01.007

    work page doi:10.1016/j.landig.2025.01.007 2025

  9. [9]

    Current progress of digital twin construction using medical imaging

    Zhao F, Wu Y, Hu M, Chang C-W, Liu R, Qiu R, et al. Current progress of digital twin construction using medical imaging. Journal of Applied Clinical Medical Physics. 2025;26(9):e70226. doi: https://doi.org/10.1002/acm2.70226

    work page doi:10.1002/acm2.70226 2025

  10. [12]

    The National Academies Collection: Reports funded by National Institutes of Health

    National Academies of Sciences E, Medicine, National Academy of E, Division on E, Life S, Division on E, et al. The National Academies Collection: Reports funded by National Institutes of Health. Foundational Research Gaps and Future Directions for Digital Twins. Washington (DC): National Academies Press (US) Copyright 2024 by the National Academy of Scie...

    work page 2024

  11. [13]

    Medical digital twins: enabling precision medicine and medical artificial intelligence

    Sadée C, Testa S, Barba T, Hartmann K, Schuessler M, Thieme A, et al. Medical digital twins: enabling precision medicine and medical artificial intelligence. The Lancet Digital Health. 2025;7(7). doi: 10.1016/j.landig.2025.02.004

    doi:10.1016/j.landig.2025.02.004 2025

  12. [14]

    Review and recommendations on deformable image registration uncertainties for radiotherapy applications

    Nenoff L, Amstutz F, Murr M, Archibald-Heeren B, Fusella M, Hussein M, et al. Review and recommendations on deformable image registration uncertainties for radiotherapy applications. Phys Med Biol. 2023;68(24). doi: 10.1088/1361-6560/ad0d8a

    work page doi:10.1088/1361-6560/ad0d8a 2023

  13. [15]

    Deformable medical image registration: a survey

    Sotiras A, Davatzikos C, Paragios N. Deformable medical image registration: a survey. IEEE Trans Med Imaging. 2013;32(7):1153–90. doi: 10.1109/tmi.2013.2265603

    work page arXiv 2013

  14. [16]

    Advancements in synthetic CT generation from MRI: A review of techniques, and trends in radiation therapy planning

    Bahloul MA, Jabeen S, Benoumhani S, Alsaleh HA, Belkhatir Z, Al-Wabil A. Advancements in synthetic CT generation from MRI: A review of techniques, and trends in radiation therapy planning. J Appl Clin Med Phys. 2024;25(11):e14499. doi: 10.1002/acm2.14499

    work page doi:10.1002/acm2.14499 2024

  15. [17]

    CBCT-Based synthetic CT image generation using conditional denoising diffusion probabilistic model

    Peng J, Qiu RLJ, Wynne JF, Chang CW, Pan S, Wang T, et al. CBCT-Based synthetic CT image generation using conditional denoising diffusion probabilistic model. Med Phys. 2024;51(3):1847–59. doi: 10.1002/mp.16704

    work page doi:10.1002/mp.16704 2024

  16. [18]

    A novel use of biomechanical model-based deformable image registration (DIR) for assessing colorectal liver metastases ablation outcomes

    Anderson BM, Lin YM, Lin EY, Cazoulat G, Gupta S, Kyle Jones A, et al. A novel use of biomechanical model-based deformable image registration (DIR) for assessing colorectal liver metastases ablation outcomes. Med Phys. 2021;48(10):6226–36. doi: 10.1002/mp.15147

    work page doi:10.1002/mp.15147 2021

  17. [19]

    Multi-atlas segmentation of biomedical images: A survey

    Iglesias JE, Sabuncu MR. Multi-atlas segmentation of biomedical images: A survey. Med Image Anal. 2015;24(1):205–19. doi: 10.1016/j.media.2015.06.012

    doi:10.1016/j.media.2015.06.012 2015

  18. [21]

    Imaging and clinical data archive for head and neck squamous cell carcinoma patients treated with radiotherapy

    Grossberg AJ, Mohamed ASR, Elhalawani H, Bennett WC, Smith KE, Nolan TS, et al. Imaging and clinical data archive for head and neck squamous cell carcinoma patients treated with radiotherapy. Scientific Data. 2018;5(1):180173. doi: 10.1038/sdata.2018.173

    work page doi:10.1038/sdata.2018.173 2018

  19. [22]

    Head and neck tumor segmentation in PET/CT: The HECKTOR challenge

    Oreiller V, Andrearczyk V, Jreige M, Boughdad S, Elhalawani H, Castelli J, et al. Head and neck tumor segmentation in PET/CT: The HECKTOR challenge. Medical Image Analysis. 2022;77:102336. doi: https://doi.org/10.1016/j.media.2021.102336

    arXiv 2022