{"id":"64604912-412c-45a0-a18d-14df29d2a3d6","arxiv_id":"2606.28513","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":5.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"HDDPM adapts DDPMs with a fixed Poisson-based heteroscedastic forward process and dose conditioning to improve quantitative recovery of low-count brain PET images across scanners.","lead":"The paper introduces HDDPM, a diffusion model variant that adds intensity-dependent Poisson noise in the forward process to better match PET imaging physics for recovering low-count brain scans. A smart generalist might read it to see how generative models can incorporate domain-specific noise properties to support lower-radiation medical imaging.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"The fixed Poisson variance module's accuracy for post-reconstruction noise in real multi-scanner data is the load-bearing assumption for the claimed physical inductive bias.","rationale":"The reader's weakest_assumption exactly isolates the same point: the Poisson module's fidelity to real (post-correction) noise distributions. Full-text details on simulation protocol would not remove this gap without an explicit empirical validation step against measured noise statistics, so the UNVERDICTED status and low confidence remain appropriate.","tokens_in":1807,"tokens_out":322,"duration_ms":17738,"concrete_test":"On the external dataset at 1% dose, extract paired low-count/standard-count reconstructions; compute empirical per-voxel variance in low-activity (<20% max) and high-activity regions; compare directly to the fixed Poisson module's predicted variance at the same activity levels. A >25% relative mismatch in low-activity regions would falsify the physical motivation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim requires that the heteroscedastic forward process (fixed Poisson-based voxel-wise variance maps) supplies a physically motivated bias by matching activity-dependent PET noise. The abstract states this module is fixed and applied to simulated low-count data (1%-50% dose) across scanners; real noise after iterative reconstruction, attenuation/scatter corrections, and multi-scanner variability is known to deviate from pure Poisson statistics. If the generated noise maps do not match empirical post-recon variance, the inductive bias reduces to an arbitrary heteroscedastic perturbation rather than a physics-derived one.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The manuscript introduces HDDPM, a heteroscedastic denoising diffusion probabilistic model for quantitative recovery of low-count brain PET images. It replaces the standard homoscedastic Gaussian forward process with a fixed Poisson-based variance module that generates voxel-wise, activity-dependent noise maps, conditioned on dose fraction. The model is evaluated on simulated low-count data (1%–50% dose levels) from three scanners using internal and external datasets, with the claim that HDDPM provides comparable overall image quality to isotropic DDPM but superior performance at the lowest (1%) external dose in reducing measurement errors across activity regions.","tokens_in":1929,"tokens_out":472,"duration_ms":30204,"significance":"If the central results hold after addressing the noise-model validation, the work would demonstrate a feasible physics-informed inductive bias for diffusion models in PET, potentially improving quantitative accuracy at ultra-low doses. This addresses a clinically relevant problem of radiation-dose reduction while preserving measurability in high- and low-activity brain regions.","major_comments":[{"comment":"Abstract: the claims that HDDPM 'significantly reduces measurement errors' and 'stood out in the lowest-dose (1%) external scans' are unsupported by any quantitative metrics, error bars, statistical tests, or ablation results, preventing assessment of whether the reported advantage is load-bearing or marginal.","section":"Abstract"},{"comment":"Variance module description (Methods): the assertion that the fixed Poisson-based variance module supplies a 'physically motivated inductive bias' by reflecting activity-dependent PET noise is load-bearing, yet the manuscript provides no empirical validation that the generated maps match post-reconstruction, post-correction variance observed in real multi-scanner data rather than only in the simulated low-count forward model.","section":"Methods (variance module)"}],"minor_comments":[{"comment":"Abstract contains line-break artifacts ('ra-diation', 'spatially') that should be cleaned for publication.","section":"Abstract"},{"comment":"The abstract omits any mention of training details, network architecture, or loss formulation; these should be summarized even at abstract level for reproducibility.","section":"Abstract"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the constructive feedback. We address the two major comments point by point below, agreeing where revisions are warranted to strengthen the presentation of results and the justification of the variance module.","responses":[{"response":"We agree that the abstract would be improved by incorporating specific quantitative support. The full manuscript reports comparative results (including regional measurement errors) for the 1% external-dose case in the Results section and associated tables/figures. We will revise the abstract to include key quantitative values (e.g., percentage error reductions with variability measures) and explicit references to the supporting analyses, ensuring the claims are directly substantiated within the abstract itself.","revision_made":"yes","referee_comment":"[Abstract] Abstract: the claims that HDDPM 'significantly reduces measurement errors' and 'stood out in the lowest-dose (1%) external scans' are unsupported by any quantitative metrics, error bars, statistical tests, or ablation results, preventing assessment of whether the reported advantage is load-bearing or marginal."},{"response":"The module implements a fixed, voxel-wise Poisson variance scaled by local activity, which follows directly from the photon-counting statistics of PET. This choice is motivated by the known non-stationary noise properties of PET prior to reconstruction. We acknowledge that the manuscript does not include a direct side-by-side comparison of the generated maps against empirically estimated post-reconstruction variances from the multi-scanner datasets. We will revise the Methods section to expand the derivation of the module, clarify its relationship to the simulation forward model, and add a brief discussion (or supplementary figure) that relates the module outputs to variance estimates obtainable from the available high-count reference data. This will make the inductive-bias claim more transparent while acknowledging the simulation-based nature of the evaluation.","revision_made":"partial","referee_comment":"[Methods (variance module)] Variance module description (Methods): the assertion that the fixed Poisson-based variance module supplies a 'physically motivated inductive bias' by reflecting activity-dependent PET noise is load-bearing, yet the manuscript provides no empirical validation that the generated maps match post-reconstruction, post-correction variance observed in real multi-scanner data rather than only in the simulated low-count forward model."}],"tokens_in":1471,"tokens_out":478,"duration_ms":30354,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The main thing to know is that this paper replaces the usual isotropic Gaussian forward process in DDPM with a heteroscedastic one driven by a fixed Poisson variance module that scales noise with local activity, adds explicit dose-fraction conditioning, and tests the result on simulated 1-50% dose brain PET from three scanners, internal plus external. They say it matches or beats standard DDPM overall and does better at the lowest dose on external data while cutting errors in both high- and low-activity regions.\n\nWhat is actually new is the concrete combination of that Poisson-based variance map with residual diffusion and dose conditioning for PET. The paper does a reasonable job of spelling out why standard DDPM noise does not match the non-stationary, activity-dependent character of PET and of running the comparison across scanners.\n\nThe soft spots are straightforward. The abstract reports no quantitative metrics, no error bars, no training details, and no ablations, so the performance claims cannot be checked. More importantly, the fixed Poisson module is presented as supplying a physically motivated bias, yet real PET noise after iterative reconstruction, attenuation, scatter correction, and multi-scanner variability is known to depart from pure Poisson statistics. Without evidence that the generated variance maps match empirical post-recon variance, the inductive bias reduces to an arbitrary heteroscedastic perturbation. That matches the stress-test concern and looks load-bearing.\n\nThis is for people building diffusion models for quantitative PET or similar modalities. A reader hunting for domain-specific noise models could extract the core idea, but the missing numbers and validation mean it does not yet support strong conclusions.\n\nI would send it for peer review rather than desk reject. The clinical goal is worthwhile and the modeling direction is coherent enough to deserve referee time, provided the full manuscript supplies the missing metrics and some check on the variance maps.","headline":"HDDPM adds a fixed Poisson variance map to make DDPM forward noising activity-dependent for low-count PET, but the abstract supplies no numbers or validation of that map against real post-recon noise.","tokens_in":2414,"tokens_out":455,"would_cite":false,"duration_ms":29659,"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":"Heteroscedastic diffusion adds activity-dependent noise to better recover quantitative values from low-count brain PET scans.","keywords":["heteroscedastic diffusion","low-count PET","quantitative recovery","Poisson noise","brain PET","denoising diffusion model"],"falsifier":"A direct comparison of the per-voxel variance predicted by the Poisson module against the empirical variance measured in real low-count PET reconstructions from the same scanners and dose levels; mismatch at multiple activity levels would undermine the claimed inductive bias.","tokens_in":2706,"feed_emoji":"","tokens_out":679,"duration_ms":18356,"temperature":0.7,"pith_summary":"The paper establishes that standard DDPMs use uniform Gaussian noise in their forward process, which does not match the non-stationary, activity-scaled noise of PET after reconstruction. HDDPM replaces this with a fixed Poisson-based variance module that generates stronger perturbations in low-activity voxels and weaker ones in high-activity voxels, while the network learns to predict the residual to standard-count images under dose conditioning. This change supplies a physically motivated bias that improves quantitative accuracy, especially in the lowest-dose regime across multiple scanners and external data. A sympathetic reader would care because accurate low-count recovery could reduce patient radiation exposure while preserving reliable measurements in both hot and cold regions.","feed_headline":"Activity-aware noise in diffusion model improves low-count PET accuracy","feed_subtitle":"Poisson variance maps place stronger corruption on low-activity regions, cutting quantitative errors especially at 1% dose across scanners.","key_machinery":"Fixed Poisson-based variance module that produces voxel-wise noise maps reflecting local activity levels.","core_discovery":"HDDPM replaces the isotropic homoscedastic forward process of standard DDPM with a heteroscedastic residual diffusion process driven by a fixed Poisson variance module; the resulting model produces voxel-wise noise maps that place stronger corruption on low-activity regions, and under explicit dose-fraction conditioning the network recovers low-to-standard-count residuals more reliably than isotropic DDPM, particularly at 1% dose on external scans.","pith_inferences":["If the Poisson module generalizes, the same architecture could be tested on other modalities whose noise also scales with local signal intensity.","The explicit conditioning on dose fraction suggests a route to training a single network that handles a continuous range of count levels rather than discrete dose bins.","Because the variance map is fixed and not learned, the approach may be easier to validate or transfer across scanners than fully learned noise models."],"forward_implications":["HDDPM reduces quantitative measurement errors in both high- and low-activity regions relative to isotropic DDPM.","Performance advantage becomes clearest on external data at the lowest simulated dose (1%).","The model remains competitive with standard DDPM on overall image quality metrics while adding the activity-aware bias.","The same heteroscedastic forward process can be applied at multiple dose fractions from 1% to 50% without retraining the variance module."],"fun_headline_variants":["HDDPM uses Poisson noise maps for low-count PET","Heteroscedastic noise improves low-count brain PET accuracy","Poisson variance aids HDDPM in quantitative PET recovery","Activity-aware diffusion model for 1% dose PET scans","HDDPM replaces homoscedastic noise in DDPM for PET"],"cache_read_input_tokens":64,"weakest_assumption_plain":"The Poisson variance module built into the forward process accurately captures the noise statistics that remain after iterative reconstruction and physical corrections on real multi-scanner data.","fun_headline_variants_meta":{"raw":{"variants":["HDDPM uses Poisson noise maps for low-count PET","Heteroscedastic noise improves low-count brain PET accuracy","Poisson variance aids HDDPM in quantitative PET recovery","Activity-aware diffusion model for 1% dose PET scans","HDDPM replaces homoscedastic noise in DDPM for PET"]},"model":"grok-4.3","cost_usd":0.006289,"raw_usage":{"total_tokens":3007,"prompt_tokens":768,"num_sources_used":0,"completion_tokens":81,"cost_in_usd_ticks":62887000,"prompt_tokens_details":{"text_tokens":768,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":2158,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":768,"tokens_out":81,"duration_ms":24343,"temperature":1.0,"reasoning_tokens":2158,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-30T01:06:46.597696+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"A direct comparison of the per-voxel variance predicted by the Poisson module against the empirical variance measured in real low-count PET reconstructions from the same scanners and dose levels; mismatch at multiple activity levels would undermine the claimed inductive bias.","supporting_citations":[],"review_version":1}