Abstract

Photon-counting CT (PCCT) provides superior image quality with higher spatial resolution and lower noise compared to conventional energy-integrating CT (EICT), but its limited clinical availability restricts large-scale research and clinical deployment. To bridge this gap, we propose MIRA, a simulated degradation-to-restoration pipeline that learns to reverse realistic acquisition artifacts in low-quality EICT by leveraging high-quality PCCT as reference. Our central insight is to explicitly model realistic acquisition degradations, transforming PCCT into clinically plausible lower-quality counterparts and learning to invert this process. The simulated degradations were validated for clinical realism by board-certified radiologists, enabling faithful supervision without requiring paired acquisitions at scale. As outcomes of this technical contribution, we: (1) train a latent diffusion model on 1,046 PCCTs, using an autoencoder first pre-trained on both these PCCTs and 405,379 EICTs from 145 hospitals to extract general CT latent features that we release for reuse in other generative medical imaging tasks; (2) construct a large-scale dataset of over 17,316 publicly available EICTs enhanced to PCCT-like quality, with radiologist-validated voxel-wise annotations of airway trees, arteries, veins, lungs, and lobes; and (3) demonstrate substantial improvements: across external data, MIRA outperforms state-of-the-art image translation methods by 15% in SSIM and 20% in PSNR, improves radiologist-rated clinical utility in reader studies, and enhances downstream top-ranking lesion detection performance, increasing sensitivity by up to 15% and F1 score by up to 10%. Our results suggest that emerging imaging advances can be systematically distilled into routine EICT using limited high-quality scans as reference. All datasets, code, and models will be publicly released.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/1477_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{LiuJun_Distilling_MICCAI2026,
        author = { Liu, Junqi AND Zhou, Xinze AND Li, Wenxuan AND Ye, Scott AND Sitek, Arkadiusz AND Yang, Xiaofeng AND Tang, Yucheng AND Xu, Daguang AND Ding, Kai AND Wang, Kang AND Yang, Yang AND Yuille, Alan L. AND Zhou, Zongwei},
        title = { { Distilling Photon-Counting CT Into Routine Chest CT Through Clinically Validated Degradation Modeling } },
        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 MIRA, a two-stage framework that distills high-quality photon-counting CT (PCCT) image characteristics into routine energy-integrating CT (EICT) scans. The core idea is to model realistic acquisition degradations (sparse view, low dose, conventional) on PCCT data, then train a latent diffusion model to reverse this degradation. The authors pre-train a large-scale autoencoder on 405,379 EICT scans and 1,046 PCCT scans, and validate their method on multiple public and private datasets. They report improvements in image quality metrics (SSIM, PSNR), radiologist-rated clinical utility, and downstream lesion detection performance.

  • 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 study addresses a clinically relevant problem by bridging the gap between advanced photon‑counting CT (PCCT) and widely available energy‑integrating CT (EICT), which is both timely and impactful. A large‑scale pre‑training strategy using over 400,000 EICT scans from 145 hospitals provides a strong foundation for learning general CT features. Clinical realism of the degradation pipeline is further validated by board‑certified radiologists, and comprehensive experiments—including multiple baselines, ablation studies, and downstream task evaluations—robustly support the claims. Finally, the planned public release of datasets, code, and models ensures reproducibility and benefits the broader research community.

  • 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.Some baselines (e. g. , NEED, SR3) are originally designed for natural images or low-dose CT denoising. The authors mention they were retrained, but details (training data, hyperparameter tuning) are missing. Please provide a table of training settings for each baseline to ensure fair comparison.

    2.The improvements in SSIM/PSNR (e. g. , +35.9%) are large, but no confidence intervals or significance tests are reported. Please add standard deviations or bootstrapped confidence intervals for all metrics.

    3.In Table 3, the “baseline” uses standard-quality CT, while “+ MIRA” uses enhanced CT. Were the detection models (MONAI, grt123) retrained from scratch on the enhanced data? Or were they fine-tuned? Please clarify. Also, why are specificity and F1 gains inconsistent (e. g. , specificity unchanged for Luna16)? A brief discussion would help.

    4.The paper rightly notes that acquiring paired EICT and PCCT is unethical due to additional radiation. However, the degradation pipeline applies simulated low-dose effects. Is there any risk that the enhanced EICT might appear “too clean” and cause radiologists to miss noise-related cues (e. g. , that a small nodule might be real rather than noise)? Please discuss.

  • 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 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.

    (5) Accept — should be accepted, independent of rebuttal

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    This paper proposes MIRA, a two-stage framework that distills high-quality photon-counting CT (PCCT) image characteristics into routine energy-integrating CT (EICT) scans. The core idea is to model realistic acquisition degradations (sparse view, low dose, conventional) on PCCT data, then train a latent diffusion model to reverse this degradation. The authors pre-train a large-scale autoencoder on 405,379 EICT scans and 1,046 PCCT scans, and validate their method on multiple public and private datasets. They report improvements in image quality metrics (SSIM, PSNR), radiologist-rated clinical utility, and downstream lesion detection performance.

  • 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 MIRA, a degradation-to-restoration framework that transfers PCCT image quality to routine EICT. PCCT scans are forward-degraded via three artifact families (sparse-view, low-dose, conventional downsampling), and a latent diffusion model, built on a continual autoencoder pretrained on 1,046 PCCTs and 405,379 EICTs from 145 hospitals, is trained to invert them. The claimed contributions are a radiologist-validated degradation simulator, a reusable continual CT autoencoder, and a 17,316-case enhanced public dataset with voxel-wise anatomical annotations. SSIM/PSNR gains over six baselines and improved downstream detection on Luna16, LNDb19, and DSB17 are reported.

  • 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 scale of the pretraining data is a major strength. The model is trained on 405,379 EICT scans collected from 145 hospitals in 19 countries, which is notably larger than the datasets used in most comparable studies on medical generative modeling. The anatomy-aware design is well motivated. The segmentation consistency loss and HU consistency loss target two of the most important risks in generative enhancement—hallucinated anatomy and distortion of quantitative values—so these choices are not just technical additions but clinically relevant safeguards. The downstream evaluation is relatively thorough. The authors do not rely on a single test set, but report detection results across three independent external cohorts, namely LUNA16, LNDb19, and DSB17, which makes the evidence for generalizability more convincing. The paper also has strong community value. If the authors release the 17,316-case annotated dataset together with the general CT latent backbone as promised, that would be a meaningful resource for the field. In addition, the leave-one-out ablation in Table 4 is clean and informative, showing that each degradation family contributes to robustness under mixed real-world settings.

  • 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.Reader study claimed in abstract, not described in body. The abstract and conclusion refer to “reader studies with board-certified radiologists,” but no protocol, reader count, sample size, rating scale, or statistical results appear in Sections 2–4.Either the protocol should be reported, or the wording revised. 2.Radiologist validation protocol under-described. Fewer than two sentences describe the validation that underpins the entire training paradigm. Number of radiologists, blinding, rating scale, and case count are unspecified. Section 1 says “radiologists” (plural) while Section 2.2 says “an experienced radiologist” (singular). 3.Downstream detection test-time condition unspecified. Table 3 retrains detectors on enhanced CT but does not state whether test images were also enhanced. If so, the comparison measures two evaluation distributions rather than two models. A cleaner control would apply the same detector to raw versus enhanced test images. 4.Continual Autoencoder under-delivered. Claimed as one of three contributions, Section 2.1 lacks a formal Memory Loss equation, training sequence, comparison against a non-continual baseline, and any catastrophic-forgetting measurement. Figure 3a is qualitative only. 5.Figure 4 quality scorer unspecified. The 0–1 scorer supporting “narrows the gap to real PCCT” has no architecture, training data, or independence statement.

  • 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 authors claimed to release the source code and/or dataset upon acceptance of the submission.

  • 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?

    This paper addresses an important and timely problem. It benefits from a competitive pretraining scale, clinically motivated loss design, a relatively clean degradation ablation, and a meaningful commitment to open-source code and data, all of which give it clear potential value to the community. However, the current manuscript still has several gaps in how key points are explained. In addition, the method overview figure should be strengthened, as the present version does not clearly highlight the main advantages claimed in the paper.

  • 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 authors collected a large-scale clinical dataset consisting of PCCT and EICT images. They propose a novel Mira strategy that explores realistic degradation pathways of high-quality PCCT images to match the quality characteristics of EICT data. This enables the generation of well-paired training data for subsequent image restoration networks, which can effectively enhance low-quality EICT images to high-quality levels. Quantitative results demonstrate that the proposed method significantly outperforms competing methods in terms of PSNR and SSIM, validating its effectiveness.

  • 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.The main strength of MIRA is that it effectively distills the superior image quality of limited-availability PCCT into widely used routine EICT scans, offering high practical value for improving clinical diagnostic accuracy. 2.The core innovation lies in its clinically validated degradation modeling, which realistically simulates real-world acquisition artifacts to generate reliable paired supervision for the restoration network.

  • 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.In Section 2.1, the authors introduce the Memory Loss module as a key component to mitigate catastrophic forgetting in the continual autoencoder. However, they provide no specific architectural details, mathematical formulation, or implementation description. If this module is adopted from prior work, the original source should be clearly cited. 2.In Section 2.2, the paper claims to employ a Latent Diffusion Model. Yet the described loss combination does not include the fundamental diffusion noise prediction loss, which is the core training objective of any diffusion model. This omission makes the actual training objective of MIRA difficult to fully understand. 3.The experimental section lacks critical implementation details for the LDM, including the sampling strategy, number of sampling steps, noise schedule, and other essential hyperparameters. These details are necessary for reproducibility and fair comparison with other diffusion-based methods.

  • Please rate the clarity and organization of this paper

    Poor

  • 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 authors claimed to release the source code and/or dataset upon acceptance of the submission.

  • 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?

    This work not only demonstrates significant effectiveness in improving the quality of clinical EICT images, but also holds substantial value in enhancing clinical diagnostic accuracy. Furthermore, the authors plan to publicly release the relevant dataset. This initiative will greatly advance further research in the medical imaging community on EICT image quality enhancement.

  • 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



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.



back to top