List of Papers Browse by Subject Areas Author List
Abstract
Generating CT Angiography (CTA) from the latent vascular structural information in Non-Contrast CT (NCCT) has significant clinical value and remains highly challenging. With the successful application of diffusion models in medical image modality conversion tasks, research on NCCT-to-CTA generation has been promoted. However, most diffusion-based methods still use random noise to model the entire CTA image, lacking explicit modeling of vascular enhancement changes caused by contrast agent injection, which limits the quality of vascular visualization in generated images. To address this limitation, we propose a Noise-Assisted Structural Difference Diffusion Model (NASD-Diff) for NCCT-to-CTA generation. In the forward process, a noise-assisted structural difference degradation operator explicitly models the attenuation of contrast-enhanced vascular signals while incorporating controllable noise for high-frequency texture modeling. In the reverse process, a dual-decoder network decouples the prediction of vascular structural differences and noise components. Furthermore, a cross-slice conditional embedding module leverages contextual information to enhance the quality of image generation, and a global perception refinement module improves the overall visual perception of the generated images. Experiments on a real clinical dataset demonstrate that NASD-Diff significantly outperforms existing methods in PSNR, SSIM, RMSE, and multiple perceptual metrics, highlighting its effectiveness and clinical potential for high-quality CTA synthesis without contrast agents.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/0935_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
N/A
Link to the Dataset(s)
N/A
BibTex
@InProceedings{XieZen_NASDDiff_MICCAI2026,
author = { Xie, Zeng AND Guo, Lijun AND Chen, Bang AND Pan, Yuning AND He, Wenming AND Ye, Xianwang AND Zhang, Rong},
title = { { NASD-Diff: A Noise-Assisted Structural Difference Diffusion Model for CT Angiography Generation from NCCT } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16890},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper proposes NASD-Diff, a diffusion-based framework for NCCT-to-CTA synthesis that explicitly models contrast-induced vascular enhancement as a structural difference term in the forward degradation process, while using controlled noise to capture high-frequency texture details. The reverse process is implemented with a decoupled structural-texture network, supplemented by cross-slice conditional embedding and a perceptual refinement module. Overall, the paper’s main contribution is to make the diffusion process more task-aligned for CTA synthesis by replacing purely noise-driven degradation with a vascular-difference-aware formulation.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
The paper addresses a clinically meaningful problem and presents a method whose core design is well aligned with that problem formulation. The overall story is coherent from Abstract to Experiments: the paper identifies the limitation of globally noise-driven diffusion for NCCT-to-CTA synthesis, proposes a structural-difference-aware degradation process, and then validates the design with both baseline comparisons and ablations. The experimental section is reasonably comprehensive for internal benchmarking, including multiple strong baselines, full-metric comparison, component ablation, and analysis of the noise coefficient. The visual results and ablation trends are also consistent with the claimed role of NASD in improving vascular clarity and texture fidelity.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
1.The evidence for clinical utility is still limited. The evaluation is based on a single carotid dataset with 10 test patients and mainly generic image-quality metrics, without reader study, downstream vascular analysis, or external validation. 2.The core novelty is concentrated in the NASD degradation idea; CSCEM, GPRM, and perceptual refinement feel more incremental, and listing “experimental validation” as a contribution weakens the novelty framing. 3.The method section leaves several implementation choices underexplained, especially the optimization/training behavior of GPRM, the exact conditioning/prediction interface of the decoupled branches, and the practical role of cross-slice inputs in a still largely 2D pipeline. 4.The claims are somewhat stronger than the evidence. Statements about “clinical potential” and “clinically acceptable limits” are not yet fully supported by the current experimental protocol. 5.For a vascular synthesis task, the evaluation lacks anatomy-aware or vessel-specific fidelity measures, which makes it difficult to judge whether the observed metric gains translate into better vascular depiction in a clinically meaningful sense.
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not mention open access to source code or data, but provides a clear and detailed description of the algorithm to ensure reproducibility.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html
N/A
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(4) Weak Accept — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
I find the paper technically coherent and reasonably well executed, with a task-aware core idea that is more thoughtful than a standard “plug diffusion into translation” approach. However, the current evidence is not yet strong enough for me to be fully convinced at the level implied by the paper’s claims. In particular, the evaluation remains mostly internal and image-metric-driven, while the task itself is vascular and clinically motivated. If the rebuttal can better clarify the methodological details and, ideally, strengthen the case for vessel-level fidelity or clinical relevance, I could see this moving upward. In its current form, though, I place it slightly above the acceptance threshold.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Review #2
- Please describe the contribution of the paper
The authors propose a framework termed NASD-Diff for generating CT Angiography (CTA) images from Non-Contrast CT (NCCT) without requiring contrast agent injection. The method aims to directly infer CTA images from the implicit vascular information contained in NCCT scans. Compared to conventional diffusion-based approaches, which typically rely on random noise to model the entire CTA image and often lack explicit modeling of vascular enhancement induced by contrast agents, the proposed framework introduces a noise-assisted structural difference degradation operator to explicitly simulate the attenuation process of contrast-enhanced vascular signals. In addition, a dual-decoder architecture is designed to decouple the prediction of vascular structural differences and noise components, thereby improving the fidelity of vascular reconstruction. The framework also incorporates a cross-slice conditional embedding module to leverage contextual information across slices, and a global perception optimization module to enhance overall visual quality. The proposed method is evaluated on a real-world dataset consisting of 492 carotid artery cases. Experimental results demonstrate that NASD-Diff outperforms nine baseline methods across multiple quantitative metrics, including PSNR, SSIM, RMSE, VIF, and GMSD. The main contributions of this work can be summarized as follows: (1) the introduction of a noise-assisted structural difference degradation operator for modeling vascular contrast attenuation; (2) a dual-decoder framework for disentangling vascular structure and noise prediction; and (3) a cross-slice conditional embedding mechanism to exploit contextual information for improved image generation quality.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
ajor Strengths Clinically important problem. A software-based strategy to directly generate CT Angiography (CTA) from Non-Contrast CT (NCCT) scans without contrast agent injection could reduce the risk for patients with iodine allergy, renal insufficiency, or multiple myeloma, aligning with MICCAI’s clinical translation priorities. Noise-assisted structural difference modeling strategy. The changes in vascular structure caused by contrast agent between NCCT and CTA were explicitly captured. Combining this strategy in the forward propagation process of the diffusion model improved the generation of vascular structure and overall texture details. Decoupling structural difference prediction and noise prediction. Optimizing vascular structure and texture details improves the structural accuracy and detail clarity of the generated CTA images. Open-source commitment. If the collected carotid artery dataset containing 492 patients and the related code from the paper could be open-sourced, it would be of great significance to the community. Clean degradation ablation. The results of ablation experiments a and b in Table 2 clearly demonstrate that each contribution proposed by the authors contributes to the overall CTA image generation.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
Most issues below are reporting gaps, not technical errors, and can be addressed in the rebuttal without new experiments. W1.A complete description of perceptual loss [Major]. In the Training and Sampling section, the authors propose using Llpips to perceive similarity from deep features, measuring the visual consistency between the generated CTA image and the real CTA image. It’s important to clarify here whether this involves overlaying the number of image channels to match the natural image, or if some other operation is performed. W2.Implementation Details [Major]. The authors do not explicitly specify in the Implementation Details whether data augmentation strategies were employed to enhance the model’s robustness and generalization capability. W3.Diffusion model coefficients [Major]. The full text does not explicitly explain the details of the diffusion model correlation coefficients used in the entire framework, including but not limited to: the number of forward and backward samples, the sampling method, etc. W4.Model parameters and running efficiency [Major]. The discussion of model parameters and computational efficiency remains insufficient. In particular, sampling time—an inherent bottleneck in diffusion-based models—is not adequately analyzed. Furthermore, the report lacks a comprehensive comparison of runtime performance with competing methods, leaving the efficiency advantages or trade-offs of the proposed approach unclear. W5.The formula details are insufficient [Moderate]. In formula 3, || ||2 represents the L2 norm or the square term, which needs to be clarified.
- Please rate the clarity and organization of this paper
Satisfactory
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not provide sufficient information for reproducibility.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html
Completeness of Figure 4: Figure 4 only shows the visual effect of the noise control factor 1.The comprehensive display of qualitative and quantitative results of components such as NASD and CSCEM in the framework can better illustrate the effectiveness of the framework.
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(3) Weak Reject — marginally below the acceptance threshold, but would not mind if accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
This paper addresses an important and clinically meaningful topic, with investigations on the training parameters of diffusion models, well-designed degradation ablation experiments, and a valuable promise of open-source release. However, five key issues (W1, W2, W3, W4, W5) require clarification rather than additional experiments. Although the authors have explored a meaningful research direction, the innovation and advantages of the proposed method are not sufficiently highlighted in the design and description of the method architecture diagram.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Review #3
- Please describe the contribution of the paper
The paper proposes NASD-Diff, a diffusion-based framework for generating CT angiography (CTA) from non-contrast CT (NCCT). It introduces a noise-assisted structural difference (NASD) degradation operator that explicitly models contrast-induced vascular enhancement via a structural difference term, and predicts this structural component and texture noise with a decoupled dual-branch network, augmented by cross-slice conditioning and a perceptual refinement module. On a private carotid dataset, the method outperforms GAN- and diffusion-based baselines on PSNR, SSIM, RMSE, VIF, and GMSD, with ablations supporting the contributions of each component.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
1.A key strength is the noise-assisted structural difference degradation operator, which reformulates the forward process using the NCCT–CTA structural difference instead of relying only on global random noise. This is interesting because it explicitly targets the vascular enhancement changes that are central to NCCT-to-CTA generation. 2.The reverse model is reasonably well designed for the problem. The Decoupled Structural-Texture Network (DeST-Net) separates structural-difference prediction from noise prediction, while the cross-slice conditional embedding module leverages adjacent-slice context and the refinement module improves perceptual quality. These components are task-motivated rather than generic additions. 3.Empirical study is fairly comprehensive.The paper compares against a broad set of GAN-, translation-, and diffusion-based baselines, reports multiple distortion and perceptual metrics, and includes both component ablations and a study of the noise control coefficient λ. The study is based on paired NCCT–CTA data from 492 patients, with explicit training/validation/test splits and registered image pairs, which strengthens the practical relevance of the experiments.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
1.The forward operator relies on m = x_CTA − x_NCCT computed from registered pairs; m will absorb misregistration and non-vascular differences, potentially biasing the learned “structural difference” away from pure vascular enhancement. 2.The use of LPIPS is clear at a high level, but the manuscript does not specify which pretrained feature backbone is used to compute the perceptual loss. 3.Although the task is clinically motivated, the evaluation is restricted to image-quality metrics and qualitative visual comparisons on a single paired carotid dataset. Stronger claims such as “clinically acceptable limits” and broader “clinical potential” are therefore somewhat ahead of the evidence currently presented, since no reader study or downstream vascular-assessment analysis is included.
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not mention open access to source code or data, but provides a clear and detailed description of the algorithm to ensure reproducibility.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html
Single-center dataset with 10-patient test set may overestimate performance; No discussion of clinically motivated constraints (e.g., HU calibration, injectate timing priors) that may stabilize vascular enhancement realism.
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(4) Weak Accept — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The paper tackles a clinically important problem with a task-specific formulation that is better aligned with NCCT-to-CTA generation than standard noise-only diffusion, and it shows consistent gains on a real paired clinical dataset. My support is not stronger because some aspects of the degradation/restoration formulation remain under-explained.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Author Feedback
N/A
Meta-Review
Meta-review #1
- Your recommendation
Provisional Accept
- Please justify your decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. In case of an invitation for rebuttal, clarify which points are important to address in the rebuttal.
This manuscript received three positive reviews, showing the superior novelty, comprehensive evaluation and well-organized structure. Therefore, a provisional accept is given based on the reviews.
