Abstract

The PI-RADS sectorization scheme serves as the standard clinical language for reporting prostate lesion locations and guiding biopsies. However, the mapping of those discrete anatomical sectors into MR image coordinates has not been standardized and currently relies on rigid geometric scripts that are non-differentiable and often anatomically incomplete. In this paper, we propose a differentiable geometric framework that learns a global consensus sectorization template in a canonical coordinate space. By parameterizing sector boundaries as learnable primitives, our model distills an anatomical prior from expert annotations via gradient descent. During inference, this global template is projected onto patient-specific zonal masks, ensuring the sectors adapt to individual morphology while strictly enforcing PI-RADS topological constraints. Evaluation on internal (N=84) and external (ProstateX, N=50) cohorts demonstrates a significant improvement in Anterior-Posterior division over state-of-the-art baselines and provides a complete 24-sector mapping by incorporating previously omitted Lateral-Central partitions. By providing a bi-directional differentiable bridge between 3D MRI clinical nomenclature, this framework enables the spatial localization of sector based biopsy results within the 3D volume. This unlocks massive clinical registries for weak-supervision tasks and establishes a foundational tool for standardized, spatially-indexed data curation. The source code and data are publicly available at https://gitlab.inria.fr/fboccara/data-driven-prostate-sectorization

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/3653_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

https://gitlab.inria.fr/fboccara/data-driven-prostate-sectorization

Link to the Dataset(s)

N/A

BibTex

@InProceedings{BocFlo_DataDriven_MICCAI2026,
        author = { Boccarato, Florencia AND Bounazou, Fahym AND Lee, Hye Lim AND Renard Penna, Raphaele AND Delingette, Hervé},
        title = { { Data-Driven Optimization of Prostate Sectorization from MRI } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 16896},
        month = {September},
        page = {pending}
}


Reviews

Review #1

  • Please describe the contribution of the paper

    The paper proposes a differentiable, data-driven framework for prostate sectorization that learns a global anatomical template from MRI and lesion annotations. By parameterising sector boundaries as learnable geometric primitives, it enables mapping between discrete clinical sector labels and 3D image space, helping ‘unlock’ weakly-labelled clinical data.

  • 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.
    • Interesting problem studied within the work; which has real clinical utility

    • Abstract and background are very clear, with motivations summarised well

    • Methods figure is very clear and easy to parse and most methodological details are clear/well-specified

    • The analysis is in-depth (reporting coverage, variance and statistical tests) over the datasets

  • 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 link to data is useful but open-sourced code to replicate the sectorization would greatly enhance reproducibility

    • The evaluation dataset is relatively small, with just over 100 samples in total; this could be expanded; however, I do acknowledge the cost of expert time and given such constraints the size is acceptable; furthermore the authors present statistical tests, which enhance the validity of the conclusions

    • In qualitative comparisons, it would be good to also include a radiologist’s secotrization as comparison

    • The method assumes accurate zonal segmentation, how do errors in the zonal segmentation propagate, this could have been investigated by artificially corrupting the zonal masks; following from this a discussion on variability in training labels would also be appreciated

    • A hyper-parameter sensitivity analysis would be extremely useful

  • 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.

    (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 work is well-written and studies an interesting and useful problem in prostate imaging. The evaluation is somewhat limited by a small dataset size but the overall presentation and evaluation warrants publication. A few revisions would enhance the work even further and I would encourage the authors to address the weaknesses section before publication.

  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [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 authors did address some of the concerns, but several still remain with respect to evaluation. I did not find the author’s explanations for some points compelling e.g., they cited space constraints for hyperparamter sensitivity analysis but results could have been presented in the repository etc. Similarly, just citing the zonal segmentation performance seems insufficient to quantify impact of segmentation noise without conducting any experiment to validate the impact of performance degradation in particular cases. Most of the points in the rebuttal were not addressed sufficiently. The paper definitely has merit but I believe given the limited number of papers that can be accepted at MICCAI, without appropriately addressing the concerns raised, this work is just shy of the threshold for a MICCAI publication. I encourage the authors to expand their evaluation in certain aspects and submit to alternative venues.



Review #2

  • Please describe the contribution of the paper

    This paper proposes a differentiable geometric framework to learn a global consensus prostate sectorization template in canonical coordinate space. The method parameterizes sector boundaries as learnable primitives and optimizes them via gradient descent from expert annotations.

  • 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 main method is novel and well scoped. Instead of using a fixed rule-based script, the paper formulates prostate sectorization as a differentiable geometric optimization problem with learnable sector boundaries in a canonical coordinate space. This is interesting because it preserves interpretability and PI-RADS structure while still allowing data-driven optimization. The latent consensus loss, which is tailored to the weak-label setting. Rather than using a standard segmentation loss, the paper optimizes sector boundaries by maximizing lesion coverage within the assigned sector while penalizing spillover into lesions assigned to other sectors. This is a thoughtful formulation for noisy sector labels and lesions that may span multiple regions. The paper includes internal holdout, temporal internal, and external ProstateX cohorts, compares against both an unoptimized initialization and a prior rule-based baseline, and reports both aggregate and rule-specific performance. The work has potential downstream impact beyond the immediate task. The bidirectional bridge between sector labels and 3D MRI space could be useful for quality control, standardized data curation, and weakly supervised learning from large clinical registries.

  • 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 main limitation is that the evaluation target is only an indirect proxy for correctness. Performance is measured by lesion coverage within the assigned sector, but this does not fully validate whether the learned 24-sector map is anatomically correct everywhere in the gland. A sectorization could improve lesion containment while still placing boundaries in clinically questionable locations outside lesion-bearing regions. This is especially important because the paper’s goal is standardized spatial mapping, not only lesion assignment. The experimental comparison is somewhat limited. Quantitatively, the paper compares mainly against PI-Init and one prior rule-based baseline, while Walter-Rittel et al. is discussed only qualitatively on page 7–8.Since Walter-Rittel et al. [7] is one of the most relevant prior automated 24-sector approaches cited by the paper, the lack of a direct quantitative comparison weakens the empirical case. The sector labels used for training and evaluation are weak and subjective, which the paper acknowledges. Only the main lesion sector is used even though lesions may span multiple sectors, and for ProstateX the external labels are a silver standard created specifically for this study, with annotators not blinded to the framework. The clinical claim is promising but not yet fully demonstrated. The paper argues that the framework can unlock large weakly labeled biopsy registries and support weak supervision, but this is a forward-looking application rather than something actually shown experimentally in the paper.

  • 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

    The main way to strengthen the paper would be to make the validation more directly reflect the claimed goal of anatomically correct sectorization. Right now, lesion coverage is a useful proxy, but it does not fully establish that the entire 24-sector map is correct outside lesion-bearing regions. A stronger quantitative comparison with the most relevant prior automated 24-sector method, Walter-Rittel et al. [7], would also improve the empirical case, since that method is currently discussed only qualitatively. I also encourage you to be slightly more explicit about the limits of the current framework. The paper already notes that the model uses a constrained global parameterization and does not yet model prostate tilt, which is a helpful acknowledgment. Expanding this discussion would make the claims feel more balanced, especially around downstream registry unlocking, which is promising but not yet demonstrated experimentally in this paper.

  • 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?

    The paper proposes a genuinely interesting and well-motivated methodological contribution: a differentiable, interpretable framework for prostate sectorization that is clinically aligned and more principled than rigid rule-based scripts. The geometric formulation and latent consensus loss are thoughtful, and the evaluation across internal, temporal, and external cohorts is a meaningful strength.

  • 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 authors addressed four core issues: evidence of downstream weak supervision, validation outside lesion-bearing regions, quantitative comparison with Walter-Rittel, and annotation quality / reproducibility. The main criticisms received responses, but some arguments are thin. I am willing to raise my score from Reject to Weak Accept.



Review #3

  • Please describe the contribution of the paper

    The paper proposes a differentiable geometric framework for prostate sectorization that learns a global sector template in a canonical space and projects it onto patient-specific anatomy. It introduces a novel LatentConsensusLoss to optimize sector boundaries from weak lesion annotations, enabling accurate mapping between MRI voxels and clinically defined prostate sectors.

  • 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 presents a novel and well-motivated formulation of prostate sectorization as a differentiable geometric problem, replacing rigid rule-based methods with a learnable and interpretable framework. The use of a canonical coordinate space and a low-dimensional parameterization enables the learning of a global anatomical prior while maintaining consistency with clinical PI-RADS definitions. The proposed objective function is well-designed for weak supervision and effectively aligns lesion locations with anatomical sectors. In addition, the evaluation is thorough, including internal, temporal, and external datasets with appropriate statistical analysis, and qualitative results demonstrate improved anatomical fidelity and consistency with clinical guidelines.

  • 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 paper has some limitations. While the method is well-designed, the performance improvements over strong rule-based baselines are moderate and not consistently significant across all anatomical divisions, with some regions showing little to no gain. This makes it difficult to fully assess the practical impact of the proposed approach. Additionally, the framework depends on accurate zonal segmentation masks, and its performance may degrade in the presence of segmentation errors, yet this dependency is not analyzed.

  • 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

    The framework relies on a global anatomical template learned across the population, which may not fully capture patient-specific anatomical variability. Furthermore, certain geometric components and parameters (e.g., fixed symmetry assumptions and temperature settings) are not extensively justified or ablated, leaving some uncertainty about their impact on performance.

  • 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 presents a novel and well-formulated approach by casting prostate sectorization as a differentiable geometric problem, offering a clear advancement over traditional rule-based methods. The method is interpretable, clinically grounded, and technically sound, with a well-designed objective function suited for weak supervision. While performance improvements are moderate in some cases, the framework consistently matches or exceeds strong baselines and provides additional value through differentiability and integration into learning pipelines. Overall, the combination of methodological novelty, clinical relevance, and solid evaluation justifies acceptance.

  • 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 authors acknowledged the segmentation dependency and justified their design choice, but robustness to segmentation errors remains an open limitation and has not been experimentally explored.



Author Feedback

Downstream weak supervision evidence (Meta-reviewer,R2). The reviewers ask whether the proposed sectorization framework translates to improved performance in downstream cancer detection. Demonstrating a full end-to-end weakly supervised cancer detection model is beyond the scope of this paper, whose contribution is the sectorization framework itself. Prior work, not published in peer-reviewed conferences, has demonstrated that lesion sector maps can be used as weak annotations for lesion localization networks (Hamzaoui’s Thesis, Chapter 5, 2023). Our contribution is improving the fidelity of that sectorization, which is a necessary prerequisite for such pipelines. To further support the data unlocking utility, we compare against the sectorization used in that prior work: our model reduces completely misaligned POST_DIV cases from 3 to 0 on ProstateX and maintains comparable rates on other datasets, while improving fully correct AP assignments on ProstateX from 23 to 35 out of 50 cases.

Validation outside lesion-bearing regions (R2). The reviewer notes that lesion coverage does not fully validate sector boundaries in lesion-free regions. We are aware of the spatial sparsity of lesion coverage: lesions are not uniformly distributed in the gland, and our dataset reflects the expected imbalance (most lesions are in the mid-gland posterior sectors). Although every sector contains at least one lesion across the training set, we have explored balancing strategies (e.g., a WeightedDataLoader) but with no meaningful impact. Instead we construct our loss function to aggregate lesions into coarser anatomical regions (e.g., anterior vs. posterior), ensuring a maximum number of lesions contribute to each parameter’s optimization. Moreover, the parameters of the geometric priors governing sector boundaries are constrained by PI-RADS guidelines and initialized accordingly, preventing anatomically unrealistic solutions. Finally, providing manually annotated prostate sectors is way too expensive to obtain, and optimizing lesion coverage on a large and diverse dataset remains the best available proxy.

Quantitative Comparison with Walter-Rittel (WR) (R2). A sector-level comparison with WR is not meaningful because their proposed sectors differ substantially from ours (unlike Hamzaoui’s): their method uses a coarser PZ grid that does not include the lateral-central division, which is one of our key contribution since it impacts the biopsy targeting gesture.

Annotation quality and reproducibility (R2,R1). The reviewers raise concerns about annotation subjectivity and reproducibility. To the best of our knowledge, this is the first public release of prostate sector information associated with a widely used public prostate MRI dataset, ProstateX. These annotations were produced under a predefined protocol, reviewed with a senior urologist, and are publicly released to facilitate reproducibility and external validation. We believe this is a valuable contribution to the community that will first foster research on prostate image analysis. Code will be made available upon acceptance.

Hyperparameter sensitivity and fixed symmetry assumptions (R1,R3). We performed ablations on the temperature parameter α and observed limited sensitivity around the selected operating point; these results were omitted due to space constraints. The symmetry assumption is inherited directly from PI-RADS topology and discussion with clinicians.

Radiologist sectorization as qualitative benchmark (R1). This would be a valuable extension in future work, as manual annotation of sectors is too expensive.

Robustness to zonal segmentation errors (R1,R3). Our method takes as input the zonal masks, and is dependent on their quality. The segmentation method we rely on achieves strong performance on ProstateX data (WG: 90.90±2.94, TZ: 86.84±4.33, PZ: 78.40±7.31 Dice), which motivated our decision to build on these masks rather than jointly optimize segmentation and sectorization.




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.

    Both the reviewers and myself acnowledge the possible utility of this approach, but I believe all of us would like to see how using these sectors helps train better cancer model compared to using lets say a rule based approach. The paper is clear and easy to read, but we remain skeptical of downstream effect on learning. Please provide some preliminary results to support that evidence.

  • 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.

    Congratulations!



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 author addressed some of the main concerns, but the response requires including further analysis on the model’s robustness.



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.

    reviews are contravserary before and after the rebuttal. Based on my own opinion, although its clinical value and technical contributions are limited if evaluated seperately, I am in favor of evaluating its scientific value as a whole.



back to top