List of Papers Browse by Subject Areas Author List
Abstract
Accurate reconstruction of the arrhythmogenic myocardial substrate—specifically the conducting isthmuses essential for ventricular tachycardia (VT) ablation—depends on capturing fine-scale scar topology from sparse, anisotropic Late Gadolinium Enhancement (LGE) MRI. Given the pronounced anisotropy (e.g., 1.2 $\times$ 1.2 $\times$ 8.0 mm) and limited slice sampling, volumetric reconstruction constitutes a highly ill-posed inverse problem, often leading to aliasing artifacts and discontinuous “staircase” geometries under discrete interpolation schemes. To address this challenge, we propose substrate reconstruction as a continuous function learning task defined on a patient-specific anatomical manifold. We introduce SHINE (Scar Heterogeneity via Implicit Neural Encoding), a framework that constructs a mathematically consistent “Digital Twin” of the diseased myocardium. The pipeline integrates automated biventricular segmentation with a Statistical Shape Model (SSM) to establish a continuous anatomical prior. To reconstruct detailed scar morphology, we employ a Bayesian SIREN architecture with sinusoidal activation functions, enabling the representation of high-frequency geometric structures and mitigating the spectral bias persistent in conventional Neural Fields. Importantly, we introduce an Entropy-Weighted Dirichlet Energy loss, which uses epistemic uncertainty to adaptively enforce smoothness in data-sparse regions while preserving high-fidelity details in data-rich regions. Evaluated on a clinical VT cohort of 40 patients, SHINE significantly outperformed discrete interpolation and weakly supervised benchmarks, achieving a Dice Similarity Coefficient of 0.93 and a volumetric error of 3.1\%. Notably, our method effectively addresses the Nyquist sampling gap, enabling recovery of sub-voxel conducting channels within a unified, topology-preserving 3D framework to support precision electrophysiology planning.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/4741_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/FluxVisionAI/SHINE
Link to the Dataset(s)
N/A
BibTex
@InProceedings{ZonDex_SHINE_MICCAI2026,
author = { Zong, Dexiang AND Leng, Shuang AND Zhao, Xiaodan AND Tan, Ru-San AND Chen, Lei AND Cheng, Jun AND Yang, Xulei AND Sia, Ching-Hui AND Chan, Mark Y. AND Hausenloy, Derek AND Zhong, Liang},
title = { { SHINE: An Entropy-Guided Digital Twin Framework for 3D Myocardial Scar Reconstruction from Sparse CMR } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16893},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
The paper presents SHINE, a framework designed to reconstruct 3D myocardial scars from sparse and anisotropic LGE-MRI data. The work addresses the ill-posed inverse problem arising from large inter-slice gaps in clinical imaging, which often lead to discontinuities and staircase artifacts in conventional reconstruction methods. To tackle this, the authors reformulate the task as a continuous function learning problem on a patient-specific anatomical manifold, integrating anatomical priors, implicit neural representations, and uncertainty-guided regularization to improve structural continuity and representation of fine-scale features.
- 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 paper addresses the reconstruction of 3D myocardial scar from sparse and anisotropic LGE-MRI, which is a well-known limitation in clinical cardiac imaging. The focus on recovering fine-scale structures relevant to ventricular tachycardia (VT) ablation highlights the potential clinical importance of the work. 2.A key strength is the reformulation of scar reconstruction from a discrete voxel-based interpolation problem to a continuous function learning task defined on a patient-specific anatomical manifold. This perspective is conceptually appealing, as it naturally avoids discretization artifacts and enables resolution-independent modeling. 3.The framework leverages anatomical information derived from Cine MRI and a statistical shape model to constrain the reconstruction. This is a meaningful design choice, as it helps regularize the ill-posed problem by embedding prior knowledge about cardiac geometry. 4.The adoption of a SIREN-based implicit neural representation is well-suited for modeling continuous spatial fields and capturing high-frequency geometric structures. This is particularly relevant for representing narrow conducting channels that may be smoothed out by conventional CNN-based approaches.
- 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 primary challenge of this task lies in reconstructing structures within large inter-slice gaps where no direct measurements are available. However, the evaluation mainly relies on slice-wise validation (e. g. , leave-one-slice-out) and indirect geometric consistency, which do not fully assess reconstruction accuracy in truly unobserved regions. As a result, it remains unclear whether the recovered fine-scale or “sub-voxel” structures correspond to anatomically accurate features. While Fig. 2 provides intuitive qualitative comparisons, the absence of a reference or ground truth makes it difficult to determine whether the reconstructed structures reflect true anatomy.
2.The figures throughout the manuscript are generally difficult to interpret and would benefit from improved clarity and more detailed annotation. For instance, Fig. 1 (Overview of the SHINE framework) lacks sufficient explanatory details and textual labeling, making it challenging to follow the data flow and understand the role of each component in the pipeline. Key elements such as inputs, intermediate representations, and model outputs are not clearly defined. In addition, Fig. 5 is not clearly explained and is not referenced in the main text, making its purpose within the paper unclear. The intended message of the figure, particularly regarding the interpretation of uncertainty and its clinical relevance, is difficult to discern from the current presentation.
3.The study uses a cohort of 40 patients. It may be insufficient to demonstrate robustness across different MRI scanner and variations in LGE acquisition protocols. In addition, important details about the dataset are missing. For example, the spatial resolution of the LGE-MRI data (both in-plane and through-plane) is not consistently or clearly reported. Given that the proposed method specifically addresses highly anisotropic data.
4.Furthermore, the manuscript lacks sufficient implementation details for key components of the method.
5.While SHINE achieves strong numerical results, its technical novelty is incremental, representing an application-level integration of established methods such as SIREN, Bayesian uncertainty estimation, and Statistical Shape Models (SSM) rather than a fundamental algorithmic breakthrough. The framework’s performance is heavily dependent on the accuracy of the initial anatomical prior; any errors in the SSM fitting or automated segmentation of atypical cardiac morphologies could propagate into significant spatial biases in scar reconstruction. (Li, Lei, et al. 2024; Zou, Rongjun, et al. 2025)
- 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 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
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.
(2) Reject — should be rejected, independent of rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
See our comments for major issues identified: 1.The primary challenge of this task lies in reconstructing structures within large inter-slice gaps where no direct measurements are available. However, the evaluation mainly relies on slice-wise validation (e.g., leave-one-slice-out) and indirect geometric consistency, which do not fully assess reconstruction accuracy in truly unobserved regions. As a result, it remains unclear whether the recovered fine-scale or “sub-voxel” structures correspond to anatomically accurate features. While Fig. 2 provides intuitive qualitative comparisons, the absence of a reference or ground truth makes it difficult to determine whether the reconstructed structures reflect true anatomy.
2.The figures throughout the manuscript are generally difficult to interpret and would benefit from improved clarity and more detailed annotation. For instance, Fig. 1 (Overview of the SHINE framework) lacks sufficient explanatory details and textual labeling, making it challenging to follow the data flow and understand the role of each component in the pipeline. Key elements such as inputs, intermediate representations, and model outputs are not clearly defined. In addition, Fig. 5 is not clearly explained and is not referenced in the main text, making its purpose within the paper unclear. The intended message of the figure, particularly regarding the interpretation of uncertainty and its clinical relevance, is difficult to discern from the current presentation.
3.The study uses a cohort of 40 patients. It may be insufficient to demonstrate robustness across different MRI scanner and variations in LGE acquisition protocols. In addition, important details about the dataset are missing. For example, the spatial resolution of the LGE-MRI data (both in-plane and through-plane) is not consistently or clearly reported. Given that the proposed method specifically addresses highly anisotropic data.
4.Furthermore, the manuscript lacks sufficient implementation details for key components of the method.
5.While SHINE achieves strong numerical results, its technical novelty is incremental, representing an application-level integration of established methods such as SIREN, Bayesian uncertainty estimation, and Statistical Shape Models (SSM) rather than a fundamental algorithmic breakthrough. The framework’s performance is heavily dependent on the accuracy of the initial anatomical prior; any errors in the SSM fitting or automated segmentation of atypical cardiac morphologies could propagate into significant spatial biases in scar reconstruction. (Li, Lei, et al. 2024; Zou, Rongjun, et al. 2025)
- 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 rebuttal improves the manuscript, now I am happy to suggest to accept this paper.
Review #2
- Please describe the contribution of the paper
This paper introduces SHINE, a novel entropy-guided digital twin framework for reconstructing high-fidelity 3D myocardial scar from sparse and anisotropic LGE-MRI. The key contribution is the reformulation of scar reconstruction as a continuous function learning problem on a patient-specific anatomical manifold using a Bayesian SIREN-based implicit neural representation. The method incorporates an entropy-weighted Dirichlet energy loss that adaptively enforces smoothness in data-sparse regions while preserving fine structural details in data-rich areas. This enables recovery of sub-voxel scar topology and conducting channels, addressing limitations of discrete interpolation and standard deep learning approaches.
- 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.
Novel continuous reconstruction formulation: The paper reformulates scar reconstruction as a continuous function learning problem on a patient-specific anatomical manifold. This is interesting because it overcomes limitations of discrete voxel interpolation and enables resolution-independent, topology-preserving reconstruction. Use of Bayesian SIREN for high-frequency detail recovery: The adoption of a Bayesian SIREN architecture effectively addresses spectral bias and allows modeling of fine-scale structures such as conducting isthmuses. This is important for capturing clinically relevant micro-structures that are typically lost in standard CNN-based approaches. Entropy-guided adaptive regularization: The proposed entropy-weighted Dirichlet energy loss is a strong contribution, as it uses uncertainty to adaptively balance smoothness and detail preservation. This helps prevent unrealistic hallucinations in data-sparse regions while maintaining fidelity where data is available. Integration of anatomical priors via digital twin modeling: The use of a Statistical Shape Model and multi-view cine MRI to construct a patient-specific anatomical manifold provides a strong geometric constraint, improving reconstruction consistency. Strong quantitative and qualitative performance: The method shows clear improvements over interpolation and deep learning baselines in both Dice score and volumetric accuracy, supported by convincing qualitative results. Uncertainty-aware outputs for clinical safety: The inclusion of predictive uncertainty maps (“safety maps”) is a valuable feature, providing additional interpretability and potential clinical utility in identifying unreliable regions.
- 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.
Limited dataset size and lack of external validation: The evaluation is conducted on a relatively small cohort (40 patients) without external or multi-center validation. This raises concerns about generalizability across different imaging protocols and populations. Potential overfitting risk due to high model flexibility: The use of high-capacity neural fields (SIREN) combined with limited training data may lead to overfitting or hallucinated structures, especially in data-sparse regions, despite the proposed regularization. Limited validation of uncertainty estimates: Although uncertainty maps are presented as “clinical safety maps,” there is no quantitative validation or expert-based evaluation showing that these uncertainty estimates reliably correspond to reconstruction errors. Clinical impact not directly demonstrated: While the method is motivated by VT ablation planning, there is no direct evaluation showing improvement in clinical decision-making or procedural outcomes. Pipeline complexity and reproducibility concerns: The framework involves multiple components (segmentation, SSM fitting, co-registration, Baayesian INR), which may complicate implementation and limit reproducibility without detailed technical documentation.
- 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
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.
(6) Strong Accept — must be accepted due to excellence
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
This paper presents highly innovative contribution to myocardial scar reconstruction. The proposed SHINE framework introduces a fundamentally strong formulation by recasting the problem as continuous function learning on a patient-specific anatomical manifold, moving beyond traditional voxel-based reconstruction. The integration of Bayesian SIREN with entropy-guided regularization is particularly compelling, as it effectively addresses both spectral bias and uncertainty in data-sparse regions.
The methodological design is elegant and well-justified, combining anatomical priors, implicit neural representations, and uncertainty-aware regularization into a coherent and principled framework. The proposed entropy-weighted Dirichlet energy loss is a key contribution, enabling adaptive smoothing while preserving fine structural details critical for identifying conducting isthmuses.
The experimental results are strong and convincing, demonstrating clear improvements over both interpolation and deep learning baselines in terms of accuracy, topology preservation, and volumetric fidelity. The ability to recover sub-voxel structures and maintain continuity across large slice gaps is particularly impressive and clinically relevant.
Furthermore, the inclusion of uncertainty-aware “safety maps” adds an important layer of interpretability and potential clinical utility. Overall, the paper is technically rigorous, novel in its formulation, and addresses a clinically significant problem with clear impact, making it highly deserving of acceptance.
- 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 paper introduces SHINE (Scar Heterogeneity via Implicit Neural Encoding), an entropy-guided digital twin framework for 3D myocardial scars reconstruction. In SHINE, reconstruction is approached as a continuous function learning problem over a patient-specific anatomical manifold generated from a Statistical Shape Model. It utilizes a Bayesian SIREN architecture (which uses sinusoidal activations) to capture high-frequency details and introduces an Entropy-Weighted Dirichlet Energy loss.
- 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 use of a Bayesian SIREN backbone is a good choice. By using periodic activations instead of traditional functions, this framework can recover the high-frequency topological details. 2.The Entropy-Weighted Dirichlet Energy loss is a good theoretical contribution. By linking regularization directly to predictive uncertainty, it may prevents the neural field from creating hallucinations without blurring the actual observed data.
- 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 framework’s geometric prior depends on biventricular segmentations and the fitting of a Statistical Shape Model (SSM). The paper does not thoroughly explore how errors, artifacts in this initial upstream step might cascade into the final digital twin reconstruction. 2.The author claim that the Entropy-Weighted Dirichlet Energy loss is vital for preventing hallucinations. However, Table 2 shows that Variant (a) which removes only the entropy regularisation achieves a volumetric error of 1.4%, which is actually superior to the proposed SHINE model’s 3.1%. The authors rely on the qualitative visuals of Figure 4 to argue that Variant (a) produces more artifacts which is insufficient. 3.The paper highlights a rapid inference speed. However, because the framework relies on fitting a patient-specific anatomical manifold and optimizing a Bayesian architecture via Monte Carlo for each patient, the training is likely to be computationally intensive. 4.The evaluation appears to be limited to a single clinical cohort with 40 patients.
- 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
How sensitive is the SHINE framework’s final scar topology to artifacts introduced during the initial segmentations and SSM fitting?
- 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 overall score is mainly influenced by the lack of analysis on error propagation from upstream segmentation and SSM fitting, inconsistencies between the claimed importance of the loss function and experimental results, insufficient discussion of computational cost versus inference efficiency, and limited evaluation on a small, single clinical cohort.
- 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
We thank the Area Chair for the summary. We also thank R1 for recognizing our continuous manifold and SIREN architecture, R2 for endorsing our entropy regularization and safety maps, and R3 for affirming our Bayesian SIREN backbone and theoretical loss contribution. We consolidated your concerns into 5 themes addressed in the revised manuscript: 1.”Cohort of 40 Patients” & “Robustness” (R1, R2, R3) We acknowledge that 40 patients is a relatively small cohort. However, given that our primary contribution is a methodological innovation, this dataset rigorously demonstrates robustness. It originates from two centers with different MRI scanners (1.5T Siemens Avanto vs. 3.0T Biograph mMR). Acquisitions were resampled to 1.2x1.2x8.0mm space. Our Leave-One-Slice-Out(LOSO) protocol evaluates SHINE against 455 physical slices covering extreme infarct sizes (4.7%-56.1%), validating robustness against equipment variations. To clinically validate this breakthrough, we secured a multi-center dataset of >500 cases across 3 centers for future work (Will add to Sec. 4). 2.”Truly Unobserved Regions” & Lack of Ground Truth (R1) Absolute 3D ground truth data, such as 3D single-breath-hold LGE-CMR, are not available in the current study. However, our data provides relatively dense spatial sampling (8-16 slices/patient). To address this, we used a LOSO cross-validation protocol. By withholding true physical LGE slices (isolated from optimization), we establish a surrogate ground truth. Evaluated against hidden slices, SHINE achieves high spatial overlap (DSC 0.93) and near-perfect fidelity (MSE 4.0x10^-5), proving reconstructed sub-voxel structures are anatomically accurate, not hallucinations (Will detail in Sec. 3.1). 3.The “1.4% Volumetric Error” Paradox (R3) Volumetric error without topological constraints is mathematically misleading. Evaluated on LOSO hidden slices, Variant (a) collapses micro-structurally with a high MSE (1.9x10^-3). As Fig. 4A shows, its seemingly superior 1.4% error is a volumetric coincidence driven by “salt-and-pepper” hallucinations: random over- and under-estimations cancel out, destroying conducting isthmus continuity. Conversely, SHINE’s entropy loss suppresses these artifacts (MSE 4.0x10^-5). Thus, our 3.1% error reflects strict topological fidelity essential for ablation planning; the 1.4% is a dangerous illusion of fragmented noise (Will clarify in Sec. 3.1). 4.”Error Propagation from Segmentation & SSM” (R1, R3) While recent works highlight upstream error cascades (Li et al. 2024; Zou et al. 2025), our statistical analysis reveals no significant correlation (r=0.20, p=0.34) between initial segmentation/SSM geometric deviations (mean 1.59+/-0.22mm) and downstream volumetric errors. Thus, these upstream errors do not propagate into spatial biases. Mechanistically, SHINE inherently decouples anatomical geometry from pathological texture learning. Its manifold alignment effectively absorbs initial geometric noise as rigid translations without distorting the underlying scar topology (Will cite/detail in Sec. 3.1). 5.”Pipeline Complexity” & “Computationally Intensive” (R1, R2, R3) To resolve clarity and reproducibility concerns, we will overhaul Figs. 1 and 5 with explicit labels, detailing Fig. 5 in a new Sec. 3.3.We will also add Sec. 2.5 (“Implementation Details”) documenting precise hyperparameters (e.g., 4-layer Bayesian SIREN, 256 units, T=50, lambda_entropy = 5 x 10^-4) and link to our public codebase. Computationally, the ~7.5 mins/case patient-specific fitting is strictly a pre-operative process, seamlessly enabling 0.54s real-time intraoperative inference on an RTX 4090 (Will discuss in Sec. 3.3). Concluding Remarks: We will expand Sec. 1 to explicitly clarify our algorithmic advancements over standard INRs, and update Sec. 4 to outline protocols for future large-scale clinical validation and impact assessment. We sincerely thank the Area Chair and reviewers for improving our work.
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.
Reviewers agreed that the paper presents an interesting framework for reconstructing 3D myocardial scar from sparse LGE-CMR. Positive aspects included the continuous implicit reconstruction formulation and the use of anatomical priors. However, the major issue is validation. R1 especially criticised the validity of validation and clarity of illustration. Other reviewers also requested clearer reporting of computational cost, implementation details, and generalizability. In the rebuttal, the authors should clarify how reconstruction accuracy is validated in missing regions, better justify the contribution of the entropy regularizer given the ablation results, and discuss robustness and generalizability to external data.
- 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 rebuttal provides reasonable clarifications of the concerns. Some limitations remain, especially the small cohort size and lack of definitive external validation or true dense 3D ground truth. However, the rebuttal addresses the major concerns, overall, the methodological contribution and clinical relevance are sufficient for acceptance, provided the final version clearly documents the validation protocol, ablation interpretation, and remaining limitations.
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.
The reviewers found the problem clinically important and appreciated the continuous reconstruction framework, anatomical priors, Bayesian SIREN design, and entropy-guided regularization. In the final version, please add the clarifications promised in the rebuttal, especially on the LOSO validation protocol, dataset and runtime details, implementation settings, the entropy-regularization ablation, and the limitations around true 3D ground truth and clinical validation.
Meta-review #3
- 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.
Post-rebuttal comments indicate that the reviewers are satisfied with the paper which promises to be an interesting contribution to MICCAI