List of Papers Browse by Subject Areas Author List
Abstract
MR-based synthetic CT (sCT) generation has enabled MR-only radiotherapy workflows for selected anatomical sites. Despite that, patients with metal implants remain largely excluded from both clinically approved workflows and academic deep learning (DL) studies due to severe MR signal voids and metal-induced artifacts. In this work, we propose a dedicated DL framework for sCT generation in underrepresented patient cohorts with metal implants, with a focus on hip implants. Specifically, we introduce: (1) an automated method to derive and embed MR-based implant attention masks; (2) physically-guided MR augmentation strategy tailored to metal-induced artifacts and CT augmentation; (3) a systematic ablation study of multiple attention mechanisms and augmentation schemes using evaluation metrics targeting implant reconstruction. Experiments on two in-house pelvic datasets and the public SynthRAD dataset highlight clinically relevant benefits of the proposed methods, including for rare implant types. The DL-generated sCTs show strong dosimetric agreement, demonstrating clinical feasibility and supporting further development of inclusive workflows for populations previously considered unsuitable for MR-only radiotherapy. The source code is available at: https://github.com/medical-physics-usz/sct-implants.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/2030_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/medical-physics-usz/sct-implants
Link to the Dataset(s)
SynthRAD2023 dataset (Dataset 3): https://zenodo.org/records/7260705
BibTex
@InProceedings{ZalNic_Attention_MICCAI2026,
author = { Zala, Nico Camillo AND Lapaeva, Mariia AND Günther, Manuel AND Banchieri, Vittoria AND Deck, Jeanette Carmen AND Sutter, Reto AND von Deuster, Constantin AND Andratschke, Nicolaus AND Guckenberger, Matthias AND Tanadini-Lang, Stephanie AND Dal Bello, Riccardo},
title = { { Attention is Matter for Inclusiveness: Generating Synthetic CT for Patients with Hip Implants } },
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
The paper proposes a deep learning framework for synthetic CT (sCT) generation from MRI in patients with hip implants, a population typically excluded from MR-only radiotherapy workflows due to severe metal-induced artifacts. The method combines implant-aware attention mechanisms derived from MR signal voids with implant-specific data augmentation strategies, including both donor-receiver augmentation and physics-guided simulation of MR signal loss. The approach is evaluated on two in-house pelvic datasets and an external dataset, using image similarity and dosimetric metrics to assess 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.
1.The focus on enabling MR-only workflows for patients with metal implants addresses an important gap in current radiotherapy practice and is of clear translational interest. 2.The use of susceptibility-based modeling to simulate MR signal voids is well-motivated and represents a thoughtful integration of domain knowledge into the learning pipeline. 3.Evaluation across in-house and public datasets strengthens the generalizability claims. 4.Inclusion of DVH-based metrics and gamma analysis is a positive step beyond purely image-based evaluation.
- 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 proposed framework combines existing components (Pix2Pix, attention mechanisms, augmentation strategies), and the overall contribution appears incremental rather than fundamentally novel. 2.The paper is difficult to follow in several sections, particularly regarding the interaction between augmentation strategies and attention mechanisms. The contribution is not cleanly isolated, making it challenging to assess which components drive performance gains. 3.Table 1 is densely structured and not clearly organized, making it challenging to extract meaningful comparisons across methods. Many reported improvements are small and within the range of standard deviations. 4.It is unclear which specific DVH metrics (e. g. , D95%, D2%, Dmean) are presented in Figure 2B and how they correspond to specific structures. The emphasis on PTV Dmean (Table 2) is not aligned with standard clinical practice in prostate radiotherapy, where D95% is a primary metric for target coverage. 5.The number of patients with implants remains small (e. g. , 20 patients in Dataset 1 and only 5 in Dataset 2), which limits the statistical robustness of the conclusions.
- 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 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.
(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?
The paper tackles a clinically meaningful problem and incorporates well-motivated components such as physics-informed augmentation and implant-aware attention. However, the overall contribution is incremental, and the presentation lacks clarity, making it difficult to fully assess the method. More importantly, the quantitative results are difficult to interpret, and the evaluation metrics do not clearly demonstrate clinically meaningful improvements. Given these limitations, the paper falls slightly below the acceptance threshold for MICCAI.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Reject
- [Post rebuttal] Please justify your final decision from above.
The rebuttal clarifies several points, particularly the intended novelty of the augmentation/attention combination and the DVH metrics included in Fig. 2B. However, my main concerns remain only partially addressed. The contribution is still largely incremental, the presentation of the quantitative and dosimetric results in the submitted manuscript is difficult to interpret, and the small number of implant cases, especially for rare implant types continues to limit the strength of the conclusions. I appreciate the authors’ clarification that PTV D95% was computed and could be added. Overall, the rebuttal improves my understanding of the work but does not sufficiently change my assessment.
Review #2
- Please describe the contribution of the paper
The paper proposes an implant-aware deep learning framework for MR-based synthetic CT generation in patients with hip metal implants, a cohort often excluded from MR-only radiotherapy workflows. Its key components are MR-derived implant attention masks, implant-specific MR/CT augmentation including a physics-guided metal artifact simulation, and an ablation of attention strategies within pix2pix for total hip replacements and intramedullary nails. The paper also evaluates clinical feasibility with image, metal-reconstruction, and dosimetric metrics across internal and public data (SynthRAD).
- 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.Clinical motivation from the inclusiveness angle (an estimated 10-15% of prostate radiotherapy patients having implants). 2.Solid solution of using metal candidate masks and physics-guided augmentation. 3.Comprehensive evaluation metrics (image similarity, DVH, Gamma).
- 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.Limited dataset size: only a handful of subjects are used in the evaluation. 2.Limited diversity on MR scanner and sequence. The datasets isare insufficient to robustly assess generalization. This raises concerns because MR intensity distributions, artifact characteristics, and metal-induced distortions can vary substantially across vendors, field strengths, and sequences 3.Incremental technical novelty: this is minor because the merit of this work is more on the clinical side. But still, the backbone (pix2pix) is outdated, and there are many newer architectures in the SynthRAD challenge.
- 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 is a well-motivated paper on a clinically important and underexplored problem. The main concern is the dataset size/diversity.
- 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
Review #3
- Please describe the contribution of the paper
The main focus of the paper is on improving the reliability of MR based synthetic CT in presence of metal implants in MR only RT planning. The authors have demonstrated MR physics-based methods to generate the metal related voids and thresholding based methods (but carefully processed to ensure air-pockets are not misconstrued as metal regions.).The derived MR implant masks are integrated into a Pix2Pix‑based sCT generator using multiple attention mechanisms (weighted losses, SPADE normalization, and mask‑guided attention). This explicitly focuses learning capacity on metal‑affected regions, improving HU accuracy and reducing dosimetric errors near implants.
- 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.Grounded method for data geneartion : A physics driven method to simulate metal related signal voids. This helps with simulating voids for different implant types (material, shapes etc). Additonally, automated method to derive implant surrogate masks directly from MR images. 2.systematic investigation of how different attention mechanisms affect metal‑affected sCT generation with Attention driven by MR‑derived implant masks 3.Demonstration of different data augmentation methods on generated sCT with failures clearly articulated. 4.Impact of sCT generation schemes on RT planning through DVH errors
- 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.Limited conceptual novelty of sCT generation. Though the authors have carefully utilized the metal mask region and adapted it within existing Pix2Pix‑style conditional GANs formulation. 2.Physics‑Guided MR Augmentation Is Simplified and Partial. Authors are refereed to this work for improved physics driven metal augmentation strategy : K.Keskin, A. R.Sanson, B. A.Hargreaves, and K. S.Nayak, “Open-Source Simulator of Imaging Near Metal at Arbitrary Magnetic Field Strengths,” Magnetic Resonance in Medicine95, no. 4 (2026): 2370–2383, https://doi.org/10.1002/mrm.70163.3.No specific solution for cases where air pockets are constructed to be metal regions. This can be significant in abdomen and pelvis regions and result in erroneous RT dose planning. The paper doesn’t offer any specific considerations for resolving this ambiguity
- 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.
(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?
1.Physics‑guided MR augmentation 2.Evaluation is clinically meaningful with dosimetry validation (DVH metrics) 3.The paper is well written, the experimental design is carefully structured (including ablation studies and external validation), and the results convincingly show improvement relative to standard sCT baselines 4.Methodological novelty is largely incremental. 5.Fundamental ambiguity in MR‑only sCT generation that is not fully resolved: air‑filled regions are occasionally reconstructed as high‑density or metal‑like regions in the synthetic CT 6.MR Physics related simulation can be further improved.
- 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 #4
- Please describe the contribution of the paper
The authors proposed to synthesize pseudo-CT from MRI for patients with metal implants. The proposed methods introduce: (1) an automated method to derive and embed MR-based implant attention mask; (2) physically-guided MR augmentation strategy tailored to metal-induced artifacts and CT augmentation; (3) a systematic ablation study of multiple attention mechanisms and augmentation schemes using evaluation metrics targeting implant reconstruction.
- 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 proposed work offers a robust and realistic approach to utilizing abundant, available healthy MRI and CT paired data for creating a data augmentation method. This method generates data for patients with prosthetic implants to support MRI-only radiotherapy. It appears versatile and can be adapted to various implant shapes and organs. Extensive experiments show improved anatomical consistency in the generated sCT, supported by approved dosimetric evaluations.
- 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.
The authors should clarify whether they are among the first to propose a paired data augmentation method for MR to CT translation involving metal implants. This clarification could help justify the absence of comparable methods, if any exist.
- 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 has provided an anonymized link to the source code, dataset, or any other dependencies.
- 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?
The paper is well written and presented. However, the main reason for acceptance is the quantitative results of the ablation study, which support enhancing sCT generation relative to baseline.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Accept
- [Post rebuttal] Please justify your final decision from above.
The proposed method for paired data augmentation for patients with metal implants is interesting, as is its application to MR-to-CT synthesis with accompanying dosimetric evaluation.
Author Feedback
Thank you for the valuable feedback. We addressed it below (Reviewer #, Weakness #): [R1W1,W2;R2W3;R3W1,W3;R4W1,limited tech novelty] While our primary focus is a clinically relevant problem, enabling more inclusive MR-only workflows for underrepresented patient cohorts with metal implants, we adopt Pix2Pix due to its proven robustness in synthetic CT generation for different treatment sites, imaging protocols and centers. Our technical contribution lies in (i) being among the first to propose a paired data augmentation method for MR to CT translation involving implants and address rare implant scenarios.(ii) We further address metal\air ambiguity (Fig.2, Dataset 1), known challenge in sCT generation, via automated MR-only based metal candidate attention masks extraction.(iii) We provide an ablation study of augmentation and attention methods, evaluated on in-house and public datasets (code and SynthRAD dataset annotation will be released). While newer architectures will be explored in future work, our results show that only combining attention and clinically inspired augmentation improves performance in challenging implant cases, showing feasibility and addressing the key challenge of data scarcity.
[R1W5;R2W1,W2,dataset size,MR protocols] Dataset scarcity for patients with metal implants is a known limitation, further compounded by their exclusion from MR-only radiotherapy workflows, meaning paired MR-CT data are often not acquired under standardized protocols. We focus on addressing this issue by proposing augmentation strategies to extend training datasets, including rare implant types and different MR protocols, and evaluate method applicability across different MR protocols via external validation on dataset 3.To support statistical interpretation to partially address limited test sample sizes, we report repeated-measure design PERMANOVA results (Fig.2,B) including effect sizes (R²), indicating measurable group effects, along p-values. We agree that larger multicenter studies will be required to confirm robustness, and we will emphasize it more strongly. In light of this limitation, we do believe this feasibility study may further encourage more structured data collection and larger multicenter efforts for underrepresented cohorts, including federated learning or open challenges such as SynthRAD.
[R3,W2,comparison of MR metal simulations] Our physics-guided (Ph) augmentation is based on [4, 6, 14, 21]. Its novelty lies in simulating metal artifacts in GRE/Dixon MR images via the sinc-product attenuation applied to real MRs for augmentation. It shares only Δf computation logic with Keskin et al., whose MSI spin-echo simulator relies on spectral binning not applicable to GRE/Dixon [our use-case], where dephasing dominates. We will clarify these distinctions in the introduction. Overall, the use-case and validation differ (Keskin et al.:phantom + 1 patient, focused on MR); our work targets MR-CT paired augmentation. Ph supports adaptation via configurable parameters in code and has potential to be applicable to different sequences [future study].
[R1,W3,W4; table, figure] DVH metrics in Fig.2B include PTV (D95%, D2%, Dmean for prostate) and OARs (D2%, Dmean for bladder&(rectum|colon)), as stated in chapter 3.2.PTV Dmean is included in Table 1 as a proxy for HU accuracy, PTV D95% could be added, as it was already computed. Statistical differences are shown in Fig. 2B (PERMANOVA: Dataset 1:*p<0.05, R²≈0.06,small effect; Dataset 2:**p<0.01, R²≈0.70,strong effect), though we acknowledge limited statistical power and overlapping standard deviations, common in sCT studies (SynthRAD). Importantly, we emphasize the reduction of DVH outliers beyond 2%, when combining augmentation and attention (Fig. 2B, Table 1). Across all datasets and implant types, mean DVH differences remained below 2%, supporting feasibility [12]. We will revise the text, Table 1, and Fig. 2 to improve clarity and highlight clinically relevant differences
Meta-Review
Meta-review #1
- Your recommendation
Invite for Rebuttal
- 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.
The paper received 4 reviews. All reviewers viewed the generation of sCT for patients with hip implants as clinically relevant and appreciated the rigor of analysis in terms of evaluating the impact of signal voids with simulated scans as well as use of dosimetric analysis to assess the impact of synthetic CT generation errors. All reviewers found the approach only moderately innovative in terms of technique and also viewed the number of scans used for analysis may be highly limited in number for statistical power as well as potentially limited generalization of the method to MR variations. Authors are encouraged to carefully address all of the reviewers’ concerns in their rebuttal.
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation.
The authors addressed all reviewers’ concerns. However, one reviewer is still concerned regarding the presentation of results including dosimetric evaluations. It would help to clarify these results better to highlight what the contributions are.
Meta-review #2
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation.
Note:
Reviewers 2 and 3 failed to submit post-rebuttal evaluations. I contacted both, and reviewer 2 answered the following: «My recommendation would be “Accept”. But I believe the meta-reviewers can take the other reviewers’ recommendations/comments into consideration, especially when they have contradicting views.» I am basing my evaluation on the available information.
Justification:
In view of the rather positive comments of the reviewers, I back up an approval with the hope that the authors will implement the required improvements in the camera-ready version. In particular, I expect a discussion on the contributions, physics-guided augmentation and dosimetric evaluation as included in the rebuttal.
Meta-review #3
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Reject
- Please justify your recommendation.
While the work addresses an important clinical problem, the submission is not sufficient for acceptance. The contribution is largely incremental, relying on existing Pix2Pix-style synthesis, attention mechanisms, and augmentation strategies, and the manuscript does not clearly isolate which components drive the reported gains. The quantitative and dosimetric results are difficult to interpret, with small improvements that may fall within standard deviations, unclear presentation of DVH metrics, and limited alignment with clinically standard endpoints such as PTV D95%. The evaluation is also weakened by the small number of implant cases, limited diversity of implant types and imaging settings, and insufficient evidence of generalization across scanners, sequences, vendors, and rare implant scenarios. Additional concerns include the simplified physics-guided artifact simulation, unresolved ambiguity between air pockets and metal regions, and limited statistical robustness. Given these issues and the lack of convincing arguments from the rebuttal, I recommend rejection.