{"id":"eef1191d-58f7-4eb7-a0ad-ccfe0bc9892e","arxiv_id":"2606.25255","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":5.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"MR-DiffuSR applies cross-modality structural swin-attention from HR T1w images to guide 3D latent diffusion super-resolution of FLAIR, with mixed-scale training and DINOv3 loss, yielding PSNR 32.46 dB and robust downstream WMH segmentation on ADNI-4.","lead":"The paper introduces MR-DiffuSR, a 3D latent diffusion model that uses high-resolution T1-weighted MRI as structural guidance to super-resolve thick-slice FLAIR scans while aiming to avoid anatomical hallucinations. A smart generalist might read it because faster, lower-quality MRI scans are common in clinics, and better reconstruction could improve diagnostic reliability without longer scan times.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"Cross-modality swin-attention requires perfect T1w-FLAIR alignment that is assumed but unverified","rationale":"The reader's weakest assumption is exactly the load-bearing precondition for the attention mechanism to function as described. Full-text methods may contain registration details that would mitigate the concern; absent those details the assumption remains the single point where the central claim is least secure.","tokens_in":1818,"tokens_out":331,"duration_ms":15800,"concrete_test":"In the methods section, locate any registration step between T1w and FLAIR (e.g., affine or deformable). If absent, apply 2-voxel and 4-voxel rigid shifts to the T1w input on a held-out ADNI subset, recompute PSNR/SSIM/LPIPS and WMH Dice at 5x and 10x downsampling; if metrics drop >5% or visual hallucinations appear, the claim does not hold under realistic conditions.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim is that cross-modality structural swin-attention derives maps from HR T1w and applies them to LR FLAIR latents, disentangling anatomy from contrast to prevent hallucinations. This transfer is valid only if the modalities are spatially registered to sub-voxel accuracy and share identical underlying anatomy. The abstract states no registration procedure, no robustness test to misalignment, and no ablation on synthetic shifts; if even modest misalignment exists (common in multi-sequence clinical scans), the attention maps impose incorrect structural constraints on FLAIR features, which can create rather than suppress hallucinations.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The manuscript introduces MR-DiffuSR, a 3D latent diffusion framework for super-resolving low-resolution (thick-slice) FLAIR MRI using high-resolution T1w structural priors. The core component is a cross-modality structural Swin-attention module that extracts attention maps from the HR T1w image and applies them to LR FLAIR latent features to disentangle anatomy from modality-specific contrast, thereby preventing hallucinations. Training incorporates a mixed-scale degradation strategy and a DINOv3-based perceptual loss; evaluation on ADNI-4 reports aggregate metrics (PSNR 32.46 dB, SSIM 0.97, LPIPS 0.07) and improved downstream white-matter hyperintensity Dice scores relative to CNN and 2D diffusion baselines.","tokens_in":1973,"tokens_out":477,"duration_ms":14513,"significance":"If the alignment assumption holds and the attention mechanism demonstrably suppresses hallucinations, the approach could meaningfully improve robustness of diffusion-based MRI super-resolution for clinical use, especially given the downstream segmentation evaluation and the mixed-scale training for variable slice thicknesses. The explicit use of a perceptual loss grounded in DINOv3 is a constructive design choice.","major_comments":[{"comment":"Abstract: The central claim that cross-modality structural Swin-attention prevents hallucinations rests on the unstated premise that T1w and FLAIR volumes are registered to sub-voxel accuracy and share identical underlying anatomy. No registration procedure, alignment verification, or robustness experiment (e.g., synthetic shifts) is described; modest misalignment would cause the transferred attention maps to impose incorrect structural constraints, potentially creating rather than suppressing hallucinations.","section":"Abstract"},{"comment":"Abstract: Reported metrics are aggregates without error bars, per-subject standard deviations, or statistical tests. No ablation isolating the contribution of the cross-modality Swin-attention module is mentioned, leaving the load-bearing architectural claim unsupported by controlled evidence.","section":"Abstract"}],"minor_comments":[{"comment":"The abstract states results 'across all downsampling factors' but does not enumerate the tested factors or confirm that the mixed-scale training distribution matches the evaluation distribution.","section":"Abstract"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the detailed and constructive review. The comments identify important gaps in the description of data assumptions and in the quantitative support for the core architectural claim. We respond to each point below and will revise the manuscript accordingly.","responses":[{"response":"We acknowledge that the manuscript does not explicitly describe the registration procedure or include robustness experiments. In the revised version we will add a dedicated preprocessing subsection specifying the registration method (affine registration via ANTs with mutual information), alignment verification (e.g., landmark-based checks and overlap metrics on segmented structures), and a new experiment that applies controlled synthetic shifts (1–3 voxels) to the T1w prior and reports the resulting change in PSNR, LPIPS, and downstream Dice to quantify sensitivity of the attention module.","revision_made":"yes","referee_comment":"[Abstract] Abstract: The central claim that cross-modality structural Swin-attention prevents hallucinations rests on the unstated premise that T1w and FLAIR volumes are registered to sub-voxel accuracy and share identical underlying anatomy. No registration procedure, alignment verification, or robustness experiment (e.g., synthetic shifts) is described; modest misalignment would cause the transferred attention maps to impose incorrect structural constraints, potentially creating rather than suppressing hallucinations."},{"response":"We agree that aggregate metrics alone are insufficient. In the revision we will report per-subject standard deviations, error bars on all tables and figures, and paired statistical tests (Wilcoxon signed-rank) against the CNN and 2D diffusion baselines. We will also add an ablation study that removes or replaces the cross-modality Swin-attention module with standard self-attention and quantifies the resulting drops in PSNR, SSIM, LPIPS, and WMH Dice scores across downsampling factors.","revision_made":"yes","referee_comment":"[Abstract] Abstract: Reported metrics are aggregates without error bars, per-subject standard deviations, or statistical tests. No ablation isolating the contribution of the cross-modality Swin-attention module is mentioned, leaving the load-bearing architectural claim unsupported by controlled evidence."}],"tokens_in":1542,"tokens_out":456,"duration_ms":18398,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The key takeaway is that MR-DiffuSR uses cross-modality Swin attention to pull structural information from high-resolution T1w scans to guide 3D latent diffusion super-resolution of thick-slice FLAIR. It reports average PSNR of 32.46 dB, SSIM 0.97, LPIPS 0.07 on ADNI-4, and keeps Dice at 0.63 for white matter hyperintensity segmentation even at 10x downsampling where baselines fall to 0.51.\n\nWhat is new is the application of this attention transfer in 3D latent space combined with mixed-scale degradation training and a DINOv3 perceptual loss. The paper does well in including a downstream task that shows the super-resolved images remain useful for lesion analysis, which is more relevant than pure image metrics alone.\n\nThe soft spots are clear. There are no error bars or variance reported, no ablation of the attention module, and no details on registration between T1w and FLAIR. The stress-test concern is on point: if the modalities are not aligned to high precision, the attention maps could force incorrect structure onto the FLAIR features and create hallucinations rather than prevent them. The abstract does not address this, so the central claim rests on an unverified assumption.\n\nThis paper is for groups already working on diffusion-based medical image enhancement, particularly those focused on multi-modal MRI. A reader who wants to see how perceptual losses and multi-scale training play out in this setting will find concrete numbers to consider.\n\nIt deserves serious peer review because the problem is practical and the results include a useful downstream evaluation, even though the evidence needs strengthening on variance and alignment robustness.","headline":"T1-guided 3D diffusion SR for FLAIR shows promising metrics but assumes perfect cross-modal alignment without verification.","tokens_in":2522,"tokens_out":413,"would_cite":false,"duration_ms":21558,"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":"High-resolution T1w images supply structural attention maps to guide 3D latent diffusion super-resolution of thick-slice FLAIR without hallucinations.","keywords":["MRI super-resolution","diffusion models","cross-modality guidance","FLAIR","T1-weighted","latent diffusion","structural attention","hallucination prevention"],"falsifier":"Performance drop or introduction of structural errors on test cases where T1w and FLAIR volumes are deliberately shifted by 1-2 voxels before super-resolution.","tokens_in":2731,"feed_emoji":"🧠","tokens_out":726,"duration_ms":16223,"temperature":0.7,"pith_summary":"The paper establishes that cross-modality structural swin-attention can transfer anatomical structure from high-resolution T1w scans to constrain low-resolution FLAIR features inside a 3D latent diffusion model. This separation of structure from contrast prevents the fabricated details that appear in standard CNN or 2D diffusion super-resolution. The approach matters because thick-slice FLAIR is routinely acquired to save scan time, yet its restored versions must preserve brain anatomy for reliable clinical use. Mixed-scale degradation training and a DINOv3 perceptual loss further allow the model to remain accurate across a range of slice thicknesses.","feed_headline":"T1w attention maps stop hallucinations in FLAIR super-resolution","feed_subtitle":"Cross-modality swin-attention in 3D latent space preserves anatomy at 10x downsampling where other methods fail.","key_machinery":"cross-modality structural swin-attention that derives structural attention maps from HR T1w and applies them to low-resolution FLAIR latent features","core_discovery":"MR-DiffuSR introduces cross-modality structural swin-attention that derives structural attention maps from the HR T1w and applies them to the low-resolution FLAIR latent features. This design disentangles anatomical structure from modality-specific contrast, effectively preventing hallucinations. The framework operates in 3D latent space, employs mixed-scale degradation to handle varying downsampling factors, and optimizes with a DINOv3-based perceptual loss to preserve high-frequency semantic details.","pith_inferences":["The same attention-transfer idea could be tested on other modality pairs such as T1w-to-T2w or FLAIR-to-PD without retraining the full diffusion backbone.","Because the model runs in latent space, it may scale to whole-brain volumes at higher isotropic resolutions than voxel-space diffusion approaches allow.","If registration between T1w and FLAIR is imperfect in real clinical data, an explicit alignment-correction step before attention transfer would be needed to keep the hallucination-prevention benefit."],"forward_implications":["Achieves average PSNR of 32.46 dB, SSIM of 0.97, and LPIPS of 0.07 across all tested downsampling factors on ADNI-4.","Maintains Dice score of 0.63 in downstream white matter hyperintensity segmentation at 10x downsampling where baselines fall to 0.51.","Remains effective at 7 mm equivalent slice thickness through mixed-scale training.","Outperforms both CNN-based and 2D diffusion super-resolution methods."],"fun_headline_variants":["T1w structural maps block FLAIR hallucinations in latent diffusion","Swin-attention transfers T1w structure to FLAIR for accurate super-resolution","Cross-modality guidance in 3D diffusion preserves FLAIR anatomy without hallucinations","3D latent diffusion with T1w priors avoids hallucinations in FLAIR super-resolution"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"T1w and FLAIR images are assumed to be perfectly aligned and to share identical underlying anatomy so attention maps transfer without misalignment artifacts.","fun_headline_variants_meta":{"raw":{"variants":["T1w structural maps block FLAIR hallucinations in latent diffusion","Swin-attention transfers T1w structure to FLAIR for accurate super-resolution","Cross-modality guidance in 3D diffusion preserves FLAIR anatomy without hallucinations","3D latent diffusion with T1w priors avoids hallucinations in FLAIR super-resolution"]},"model":"grok-4.3","cost_usd":0.006579,"raw_usage":{"total_tokens":3129,"prompt_tokens":780,"num_sources_used":0,"completion_tokens":80,"cost_in_usd_ticks":65787000,"prompt_tokens_details":{"text_tokens":780,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":2269,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":780,"tokens_out":80,"duration_ms":16737,"temperature":1.0,"reasoning_tokens":2269,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-25T21:04:04.654746+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"Performance drop or introduction of structural errors on test cases where T1w and FLAIR volumes are deliberately shifted by 1-2 voxels before super-resolution.","supporting_citations":[],"review_version":1}