{"id":"d2270105-cfe0-4659-be5d-75835168b637","arxiv_id":"2606.11012","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"IMPACT-DoseAcc propagates DIR uncertainty via Bayesian and ensemble segmentation models into probabilistic dose-volume histograms for cumulative dose estimation in CBCT-guided oART for LACC.","lead":"This paper introduces IMPACT-DoseAcc, a framework that estimates and propagates uncertainty from deformable image registration into cumulative radiation dose calculations for daily-adapted cervical cancer treatments using CBCT scans. Radiation oncologists and medical physicists may use it to better interpret accumulated doses when patient anatomy changes between fractions.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"Exclusion of vCT-generation uncertainty may leave dominant anatomical sources unmodeled for CBCT-based dose accumulation","rationale":"The reader's weakest_assumption directly identifies the same scoping limitation. The abstract supplies modest internal checks (Pearson 0.63–0.66, 96.3 % coverage) that support the DIR-only propagation step, but the claim language about \"anatomical variations\" and \"cumulative dose under anatomical variations\" is broader than the implemented scope. This is therefore a genuine boundary condition rather than an internal inconsistency; the concrete_test above would falsify or confirm whether the boundary matters for the headline result.","tokens_in":1799,"tokens_out":414,"duration_ms":13439,"concrete_test":"On the same nine-patient LACC cohort, recompute the ensemble DIR uncertainty maps after adding a simple vCT perturbation model (e.g., ±50 HU Gaussian noise in high-gradient regions plus bladder/rectum density shifts drawn from literature CBCT error distributions); compare the resulting pDVH coverage percentages and fraction-to-fraction variance against the DIR-only results. A >15 % change in coverage or variance indicates the excluded component is load-bearing.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim is that IMPACT-DoseAcc propagates registration-driven uncertainty to improve interpretation of accumulated dose under anatomical variations. The manuscript explicitly scopes the framework to DIR uncertainty only (\"without modeling vCT-generation uncertainty\") and tests two segmentation-based strategies within IMPACT-Reg. In the CBCT-to-vCT pipeline for LACC, however, vCT generation itself introduces substantial uncertainties (HU calibration, scatter, density mapping) that directly affect the dose recalculation step before any warping occurs. If these are comparable to or larger than DIR errors, the voxel-wise uncertainty maps and resulting pDVHs capture only a subset of the total variability, so the reported calibration (96.3 % coverage) and correlation with surface distance do not demonstrate that the propagated uncertainty is sufficient for the broader claim.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The manuscript introduces IMPACT-DoseAcc, a framework for propagating DIR uncertainty through dose warping and accumulation to produce probabilistic DVHs in CBCT-guided online adaptive radiotherapy for locally advanced cervical cancer. It tests two segmentation-based DIR uncertainty strategies (Bayesian and ensemble) on nine retrospective patients, reports Pearson correlations of 0.63-0.66 between uncertainty and surface distance, 96.3% pDVH coverage, and provides 3D Slicer integration.","tokens_in":1981,"tokens_out":485,"duration_ms":22762,"significance":"If the calibration and correlation results hold under broader validation, the framework offers a practical, modality-agnostic approach to quantifying registration-driven uncertainty in cumulative dose metrics, which could support more informed clinical decisions in adaptive workflows. The explicit 3D Slicer integration is a clear strength for reproducibility.","major_comments":[{"comment":"Material and Methods: The framework is explicitly scoped to DIR uncertainty only (\"without modeling vCT-generation uncertainty\"), yet the CBCT-to-vCT pipeline for LACC involves vCT generation steps (HU calibration, scatter correction, density mapping) that precede dose recalculation and warping. If these sources are comparable to or larger than DIR errors, the voxel-wise uncertainty maps and resulting 96.3% pDVH coverage capture only a subset of total anatomical variability, weakening the claim that propagation improves interpretation of accumulated dose under anatomical variations.","section":"Material and Methods"},{"comment":"Results: The reported Pearson correlations (0.63 for CTVt, 0.66 for bladder) and pDVH coverage of 96.3 +/- 3.9% are derived from an internal cohort of only nine patients, with no comparison to a non-uncertainty baseline dose accumulation method and no external validation cohort. This leaves the evidence for calibration and improved interpretation preliminary and limits assessment of whether the uncertainty estimates add value beyond standard practice.","section":"Results"}],"minor_comments":[{"comment":"Abstract and Results: Clarify whether the +/- 3.9% on pDVH coverage refers to a figure that lacks visible error bars, and ensure consistency between text and any accompanying figures.","section":"Abstract"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the constructive comments. We provide point-by-point responses to the major comments below.","responses":[{"response":"We agree that the manuscript explicitly limits the scope to DIR uncertainty propagation, as stated in the Methods section. vCT-generation steps introduce additional uncertainties that are not modeled here. The framework is designed to isolate and propagate registration uncertainty; we will revise the Discussion to clarify that the reported pDVH coverage reflects only this component and that total anatomical variability may be larger, with a note on potential future extensions.","revision_made":"yes","referee_comment":"[Material and Methods] Material and Methods: The framework is explicitly scoped to DIR uncertainty only (\"without modeling vCT-generation uncertainty\"), yet the CBCT-to-vCT pipeline for LACC involves vCT generation steps (HU calibration, scatter correction, density mapping) that precede dose recalculation and warping. If these sources are comparable to or larger than DIR errors, the voxel-wise uncertainty maps and resulting 96.3% pDVH coverage capture only a subset of total anatomical variability, weakening the claim that propagation improves interpretation of accumulated dose under anatomical variations."},{"response":"The study is indeed limited to a retrospective internal cohort of nine patients, with results presented as a proof-of-concept. No external validation or baseline comparison was performed. We will add an explicit limitations paragraph acknowledging the preliminary nature of the calibration metrics and the need for larger multi-center validation to assess added clinical value.","revision_made":"partial","referee_comment":"[Results] Results: The reported Pearson correlations (0.63 for CTVt, 0.66 for bladder) and pDVH coverage of 96.3 +/- 3.9% are derived from an internal cohort of only nine patients, with no comparison to a non-uncertainty baseline dose accumulation method and no external validation cohort. This leaves the evidence for calibration and improved interpretation preliminary and limits assessment of whether the uncertainty estimates add value beyond standard practice."}],"tokens_in":1502,"tokens_out":474,"duration_ms":15907,"standing_objections":["External validation on an independent cohort and direct comparison against a non-uncertainty baseline accumulation method cannot be provided without new data collection and experiments outside the current retrospective study."]},"desk_editor":{"model":"grok-4.3","letter":"The core deliverable is IMPACT-DoseAcc, which takes two segmentation-based uncertainty estimates (Bayesian single model and ensemble), turns them into voxel-wise maps, warps the dose, and produces probabilistic DVHs. On nine retrospective LACC patients the ensemble version correlates with surface distance at 0.63–0.66 and reaches 96.3 % coverage on the pDVHs. The 3DSlicer module is a concrete addition that lets others run the same steps.\n\nThe work is new in its specific end-to-end combination for this disease site and in the weighting step that stabilizes the accumulated metrics across fractions. The correlations and coverage numbers are reported directly from the data rather than derived from the same fitted parameters, so there is no obvious circularity.\n\nThe main limitation is the explicit scope: the framework stops at DIR uncertainty and does not model vCT generation (HU mapping, scatter, density assignment). In CBCT-guided workflows those steps can easily be as large as registration error, so the reported calibration only covers part of the total variability. Nine patients, no non-uncertainty baseline, and no external cohort keep the evidence preliminary. The ±3.9 % on coverage is given, but the rest of the results lack error bars or statistical comparison.\n\nThe paper is aimed at physicists and dosimetrists already working on adaptive cervical radiotherapy who need a practical way to attach uncertainty to accumulated dose. Readers outside that niche will find the methods clear but the clinical payoff modest.\n\nIt is worth sending to peer review. The implementation is reproducible enough and the problem is real; referees can push on the missing vCT term and the small cohort without the work being fundamentally broken.","headline":"The paper gives a usable pipeline for pushing DIR uncertainty into pDVH estimates on CBCT data for cervical oART, but the decision to ignore vCT-generation errors narrows what the calibration numbers actually show.","tokens_in":2511,"tokens_out":433,"would_cite":false,"duration_ms":15137,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"The uncertainty-aware dose accumulation framework propagates DIR uncertainty through dose warping to produce probabilistic DVHs for adaptive radiotherapy.","keywords":["adaptive radiotherapy","dose accumulation","uncertainty estimation","deformable image registration","cervical cancer","CBCT","probabilistic DVH"],"falsifier":"A finding that probabilistic DVH coverage falls well below 96 percent or that voxel-wise uncertainty shows no correlation with measured surface distances on warped contours would indicate the propagation step does not reliably represent the uncertainty.","tokens_in":2720,"feed_emoji":"📊","tokens_out":605,"duration_ms":24891,"temperature":0.7,"pith_summary":"The paper introduces a framework to estimate uncertainty arising from deformable image registration and propagate it into cumulative radiation dose estimates during online adaptive radiotherapy. Anatomical changes between daily fractions make it difficult to determine the actual total dose received by targets and organs at risk. Two strategies for quantifying DIR uncertainty, one Bayesian segmentation-guided and one based on an ensemble of models, generate voxel-wise uncertainty maps that are carried forward through dose recalculation and accumulation on virtual CTs derived from CBCT scans. The result is a set of probabilistic dose-volume histograms whose calibration is checked against geometric error measures in nine patients with locally advanced cervical cancer. A weighting step based on anatomical variability is used to combine results across treatment fractions.","feed_headline":"Framework propagates DIR uncertainty into probabilistic dose metrics","feed_subtitle":"IMPACT-DoseAcc turns registration variability into calibrated pDVHs for CBCT-guided cervical cancer ART","key_machinery":"Voxel-wise uncertainty maps generated by Bayesian or ensemble DIR strategies, propagated through dose warping and accumulation steps to yield probabilistic dose-volume histograms with anatomical-variability weighting.","core_discovery":"The framework focuses on uncertainty from DIR without modeling vCT-generation uncertainty. Two DIR uncertainty strategies were tested: a Bayesian segmentation-guided approach using one probabilistic model to quantify anatomical uncertainty, and an ensemble of segmentation models targeting structures to capture epistemic variability. Voxel-wise uncertainty maps were propagated through dose warping and accumulation to generate probabilistic dose-volume histograms. Ensemble uncertainty was quantified from voxel-wise standard deviation across deformation fields, and geometric error was assessed using surface distance between warped and validated contours.","pith_inferences":["The same propagation steps could be tested on other disease sites where daily anatomical variation affects cumulative dose interpretation.","Because vCT-generation uncertainty is left out, the reported uncertainty ranges may understate the full variability present in clinical use.","If the uncertainty maps prove reliable, they could be used to set thresholds for triggering plan adaptation when accumulated dose confidence intervals become too wide."],"forward_implications":["Ensemble DIR uncertainty correlated with geometric error at Pearson coefficients of 0.63 for CTVt and 0.66 for bladder.","Probabilistic DVHs for CTVt achieved 96.3 plus or minus 3.9 percent coverage.","Anatomical-variability weighting stabilized dose estimates across fractions and organs.","The approach integrates with 3DSlicer to support reproducible workflows."],"fun_headline_variants":["DIR uncertainty mapped to pDVHs in cervical oART","Propagating DIR uncertainty into cumulative dose metrics","Bayesian and ensemble DIR uncertainty for ART dose","Calibration of pDVHs via DIR uncertainty propagation","Ensemble models quantify DIR uncertainty in dose warping"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"The two tested DIR uncertainty strategies sufficiently capture the dominant sources of anatomical uncertainty for dose accumulation.","fun_headline_variants_meta":{"raw":{"variants":["DIR uncertainty mapped to pDVHs in cervical oART","Propagating DIR uncertainty into cumulative dose metrics","Bayesian and ensemble DIR uncertainty for ART dose","Calibration of pDVHs via DIR uncertainty propagation","Ensemble models quantify DIR uncertainty in dose warping"]},"model":"grok-4.3","cost_usd":0.00465,"raw_usage":{"total_tokens":2351,"prompt_tokens":767,"num_sources_used":0,"completion_tokens":69,"cost_in_usd_ticks":46499500,"prompt_tokens_details":{"text_tokens":767,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":1515,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":767,"tokens_out":69,"duration_ms":14064,"temperature":1.0,"reasoning_tokens":1515,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-27T13:16:02.236420+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"A finding that probabilistic DVH coverage falls well below 96 percent or that voxel-wise uncertainty shows no correlation with measured surface distances on warped contours would indicate the propagation step does not reliably represent the uncertainty.","supporting_citations":[],"review_version":1}