{"id":"c8da7cfb-5aad-425a-b14e-e79793c070c4","arxiv_id":"2606.23100","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":4.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"PRC-TP combines an nnFormer network trained on Monte Carlo simulations with Model-consistent Texture Re-Injection to correct positron range effects in 82Rb PET while restoring acquisition-consistent texture.","lead":"The paper introduces PRC-TP, a neural network framework to correct positron range blurring in PET imaging while using a separate model to restore realistic texture. This targets a known resolution limit for high-energy emitters like 82Rb and aims to keep images suitable for texture analysis.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"Auxiliary Noise2Noise isolation of nnFormer smoothing for MTRI lacks explicit validation against Monte Carlo noise statistics","rationale":"The reader's weakest_assumption is precisely the load-bearing step; the abstract-only review correctly flags it as unverified. No other internal inconsistency is visible from the given claims, but the decoupling mechanism must hold for the texture-preservation results to be physically meaningful rather than post-hoc.","tokens_in":1826,"tokens_out":348,"duration_ms":7538,"concrete_test":"On the held-out MC test set, compute the voxel-wise difference map between nnFormer output and GT; feed the nnFormer output through the trained Noise2Noise model and compare the resulting residual power spectrum (or local variance in uniform ROIs) against the difference map. If the spectra differ by >10% in the 0.1-0.5 cycles/mm band, the MTRI texture re-injection deviates from acquisition-consistent statistics.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim requires that the auxiliary Noise2Noise model exactly captures only the deterministic smoothing induced by the main nnFormer (trained on patient-derived MC simulations) so that MTRI can re-inject texture whose amplitude and statistics match the original acquisition physics. The abstract reports near-unity global texture amplitude (0.997 +/- 0.011) after MTRI but provides no quantitative check that the estimated smoothing residual is free of residual PR effects, anatomy-dependent biases, or mismatches with the underlying Poisson noise model. If this isolation is incomplete, the re-injected texture could be inconsistent with the GT references used for the 98.96-99.04% contrast recovery figures.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The manuscript proposes PRC-TP, a positron range correction framework for 82Rb PET that trains an nnFormer network on patient-derived Monte Carlo simulations to map PR-degraded images to PR-free references (using attenuation maps as context), then uses an auxiliary Noise2Noise model to estimate and remove the network's smoothing effect before applying Model-consistent Texture Re-Injection (MTRI) to restore acquisition-consistent texture. Quantitative results on simulated patients report contrast recovery of 98.96-99.04% with near-unity global texture amplitude (0.997 +/- 0.011) after MTRI, plus improved radiomics agreement; a single clinical case shows consistent trends in contrast ratio and texture restoration.","tokens_in":1966,"tokens_out":596,"duration_ms":17318,"significance":"If the auxiliary model isolation and MTRI re-injection are shown to be physics-consistent, the framework offers a practical route to resolution recovery in high-energy positron PET while addressing the common side-effect of over-smoothing that affects texture analysis and visual realism; the use of patient-derived MC simulations for training and the function-based MTRI formulation are explicit strengths that support reproducibility of the texture amplitude result.","major_comments":[{"comment":"Methods (auxiliary Noise2Noise and MTRI sections): the central claim that MTRI re-injects texture whose amplitude and statistics match the original acquisition physics rests on the unvalidated assumption that the Noise2Noise model isolates only the deterministic smoothing induced by nnFormer; no quantitative comparison of the estimated residual to Monte Carlo Poisson noise statistics, PR-free ground truth, or anatomy-dependent bias is reported, so the reported 0.997 +/- 0.011 texture amplitude agreement does not yet confirm consistency with the underlying physics.","section":"Methods: auxiliary Noise2Noise and MTRI"},{"comment":"Results (simulated patients paragraph): the contrast recovery figures (98.96-99.04%) and CNR/noise restoration claims are presented without error bars, ablation studies on MTRI hyperparameters, or details on training/validation splits and Monte Carlo variance, which are load-bearing for assessing whether the preservation is robust rather than an artifact of the simulation setup.","section":"Results: simulated patients"}],"minor_comments":[{"comment":"Abstract and Results: the single clinical case is described only qualitatively ('trends consistent'); adding at least basic quantitative metrics with comparison to the simulation protocol would strengthen the generalizability statement.","section":"Abstract and Results: clinical evaluation"},{"comment":"Notation: the distinction between the 'function-based MTRI formulation' and any alternative formulations is referenced but not defined with an equation; adding the explicit functional form would clarify how global texture amplitude is computed.","section":"Methods: MTRI"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for their constructive comments on our manuscript. We address each major comment point by point below, agreeing where revisions are warranted to strengthen the presentation of physics consistency and robustness.","responses":[{"response":"We acknowledge the referee's point that the manuscript lacks an explicit quantitative validation of the Noise2Noise residual against Monte Carlo Poisson statistics or anatomy-dependent bias. The MTRI approach is formulated to extract and re-inject texture based on the auxiliary model's estimate of nnFormer-induced smoothing, with the reported amplitude agreement measured directly against PR-free ground truth. To address the concern, we will add a dedicated validation subsection (or supplementary figure) comparing the estimated residual noise power spectrum and amplitude to the known Poisson statistics from the patient-derived Monte Carlo simulations used in training.","revision_made":"yes","referee_comment":"Methods (auxiliary Noise2Noise and MTRI sections): the central claim that MTRI re-injects texture whose amplitude and statistics match the original acquisition physics rests on the unvalidated assumption that the Noise2Noise model isolates only the deterministic smoothing induced by nnFormer; no quantitative comparison of the estimated residual to Monte Carlo Poisson noise statistics, PR-free ground truth, or anatomy-dependent bias is reported, so the reported 0.997 +/- 0.011 texture amplitude agreement does not yet confirm consistency with the underlying physics."},{"response":"We agree that error bars, ablation studies, and explicit details on data splits and simulation variance are necessary to demonstrate robustness. The reported contrast recovery range (98.96-99.04%) summarizes results across the simulated patient cohort; we will revise the text and figures to include mean values with standard deviations and error bars. Expanded Methods text will specify the patient-wise training/validation split and Monte Carlo variance (derived from the 10^8 decay simulations). Ablation results on MTRI hyperparameters (e.g., scaling factor) will be added to the supplementary material.","revision_made":"yes","referee_comment":"Results (simulated patients paragraph): the contrast recovery figures (98.96-99.04%) and CNR/noise restoration claims are presented without error bars, ablation studies on MTRI hyperparameters, or details on training/validation splits and Monte Carlo variance, which are load-bearing for assessing whether the preservation is robust rather than an artifact of the simulation setup."}],"tokens_in":1547,"tokens_out":504,"duration_ms":20341,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The core point is that PRC-TP trains an nnFormer on Monte Carlo 82Rb data to recover resolution, then uses an auxiliary Noise2Noise model to estimate the smoothing it introduced and re-injects texture via MTRI so the output keeps realistic noise properties.\n\nWhat is new is the explicit decoupling: the main network handles the deterministic blur correction while MTRI handles stochastic texture restoration. They report contrast recovery at 98.96-99.04% of ground truth in simulations and global texture amplitude at 0.997 ± 0.011 after re-injection, plus better radiomics feature agreement.\n\nThe work does a reasonable job showing those quantitative trends and noting that a single clinical 82Rb case follows the same pattern, including a contrast-ratio increase close to the simulated value.\n\nThe soft spot is the untested assumption that the Noise2Noise model cleanly isolates only the nnFormer smoothing without residual positron-range effects or mismatches to the underlying Poisson statistics. The abstract gives no ablation on that isolation, no error bars, and no direct comparison of the estimated residual to the Monte Carlo noise model. Training and evaluation both sit inside the same simulation framework, so the clinical case is too thin to provide independent grounding.\n\nThis is for PET reconstruction researchers who work with high-energy emitters and care about downstream texture or radiomics use. A reader looking for a practical two-stage correction method will find concrete numbers to examine.\n\nIt deserves a serious referee because the idea is clearly stated and the simulation metrics are specific, even though the auxiliary model will need tighter checks.","headline":"The paper splits positron range correction into an nnFormer step plus a separate MTRI texture re-injection using Noise2Noise, but the isolation step lacks direct validation against the simulation noise model.","tokens_in":2466,"tokens_out":402,"would_cite":false,"duration_ms":19215,"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":"A neural framework corrects positron range blurring in PET while re-injecting matching texture via an auxiliary model.","keywords":["positron range correction","texture preservation","PET imaging","82Rb","neural network","MTRI","Monte Carlo simulation","radiomics"],"falsifier":"A head-to-head comparison on real clinical 82Rb patient scans that measures texture amplitude or radiomics features against an independent ground-truth reference and finds deviation beyond the reported 0.997 plus or minus 0.011 agreement.","tokens_in":2755,"feed_emoji":"🩺","tokens_out":721,"duration_ms":19038,"temperature":0.7,"pith_summary":"The paper establishes a positron range correction framework called PRC-TP that separates resolution recovery from texture restoration in 82Rb PET imaging. A main nnFormer network maps PR-degraded images to PR-free references using attenuation maps, while an auxiliary Noise2Noise model estimates the over-smoothing so that Model-consistent Texture Re-Injection can transfer the original texture back. This matters for clinical use because uncorrected PR blurring reduces contrast and spill-out at tissue boundaries, and standard corrections often erase the stochastic texture needed for realistic appearance and radiomics. In Monte Carlo patient simulations the method recovers 98.96-99.04 percent of ground-truth contrast while returning global texture amplitude to 0.997 plus or minus 0.011.","feed_headline":"Neural framework corrects PET positron blurring while restoring texture","feed_subtitle":"PRC-TP reaches 99 percent contrast recovery and 0.997 texture amplitude match to ground truth in 82Rb simulations.","key_machinery":"Model-consistent Texture Re-Injection (MTRI), which isolates the smoothing induced by the main resolution-recovery network and transfers the original noise texture back to the corrected output to maintain acquisition-consistent statistics.","core_discovery":"PRC-TP decouples deterministic resolution recovery performed by an nnFormer network trained on patient-derived Monte Carlo simulations from stochastic texture restoration performed by Model-consistent Texture Re-Injection derived from an auxiliary Noise2Noise estimate; the combined pipeline restores contrast recovery to 98.96-99.04 percent of ground truth, returns noise and CNR closer to reference values, and achieves near-unity global texture amplitude agreement of 0.997 plus or minus 0.011 while reducing input bias.","pith_inferences":["The same separation of deterministic recovery from stochastic texture could be tested on other resolution-degrading effects such as motion or partial-volume correction.","If the MTRI step proves robust across scanners, it may reduce the need for scanner-specific texture post-processing in multi-center studies.","Extending the Monte Carlo training set to include measured rather than simulated patient data would directly test whether the reported agreement holds outside the simulation domain."],"forward_implications":["Radiomics features from texture-sensitive families show improved numerical agreement with ground-truth values.","Clinical 82Rb evaluations exhibit contrast-ratio increases comparable to simulation results together with restored texture.","The same decoupled correction-plus-re-injection structure can be applied to other high-energy positron emitters.","Noise and CNR values are returned closer to the reference distribution than resolution recovery alone."],"fun_headline_variants":["PRC-TP corrects PET positron blurring with texture preservation","nnFormer-based method restores contrast and texture in 82Rb PET","Texture re-injection preserves amplitude in PET resolution recovery","PRC-TP achieves 99 percent GT contrast with 0.997 texture match"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"The auxiliary Noise2Noise model accurately isolates only the smoothing caused by the main network without adding new biases or mismatches to the underlying acquisition physics.","fun_headline_variants_meta":{"raw":{"variants":["PRC-TP corrects PET positron blurring with texture preservation","nnFormer-based method restores contrast and texture in 82Rb PET","Texture re-injection preserves amplitude in PET resolution recovery","PRC-TP achieves 99 percent GT contrast with 0.997 texture match"]},"model":"grok-4.3","cost_usd":0.006862,"raw_usage":{"total_tokens":3236,"prompt_tokens":767,"num_sources_used":0,"completion_tokens":69,"cost_in_usd_ticks":68624500,"prompt_tokens_details":{"text_tokens":767,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":2400,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":767,"tokens_out":69,"duration_ms":29613,"temperature":1.0,"reasoning_tokens":2400,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-26T05:59:04.860992+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"A head-to-head comparison on real clinical 82Rb patient scans that measures texture amplitude or radiomics features against an independent ground-truth reference and finds deviation beyond the reported 0.997 plus or minus 0.011 agreement.","supporting_citations":[],"review_version":1}