List of Papers Browse by Subject Areas Author List
Abstract
Brachytherapy is an evidence-based, well-established treatment for intracorporeal tumors. A high-quality preoperative planning scheme is critical to ensuring optimal intraoperative efficacy. However, existing preoperative planning approaches often suffer from poor reproducibility, suboptimal dose distributions, and labor-intensive workflows. To address these limitations, we propose SeedPro, an automatic and fine-grained preoperative planning framework for brachytherapy that efficiently produces expert-level treatment plans with reduced computational costs, driven by hierarchical reinforcement learning agents. Experiments show that SeedPro achieves 12–20% reduction in OAR exposure and reduces puncture numbers by 25% compared to SOTA on both internal and external datasets. This work establishes a practical paradigm for intelligent and precision-oriented brachytherapy. The source code is available at https://github.com/Haitao-Lee/SeedPro.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/3635_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/Haitao-Lee/SeedPro.git
Link to the Dataset(s)
MSD dataset: http://medicaldecathlon.com
BibTex
@InProceedings{LiHai_SeedPro_MICCAI2026,
author = { Li, Haitao AND Liu, Jiaxuan AND Huang, Wei AND Wang, Zhongmin AND Chen, Xiaojun},
title = { { SeedPro: Place Brachytherapy Seeds Like Expert Clinicians via Hierarchical Reinforcement Agents } },
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
This paper presents a hierarchical reinforcement learning agent approach to pre-operative planning for low-dose-rate brachytherapy applications with a key novel feature being the prioritization of minimal needle punctures, reducing procedural complexity. The proposed approach is tested on a local data set with a mix of anatomical sites and an external pancreatic dataset, showing improved performance on all described metrics relative to a comparative state-of-the-art algorithm that has already been validated against manual/expert plans.
- 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 key strength of this paper is the emphasis on mimicking current clinical decision-making strategies in the proposed approach, overcoming a major limitation of existing strategies. In particular, the inclusion of number of needle punctures as a key metric is a very important factor for clinical utility.
- The results presented seem promising, showing improvements on all reported metrics over state-of-the-art algorithms. Studies included both comparisons with internal and external datasets and analyses of robustness, sensitivity, and ablation. 3.The paper is well-written and detailed making it highly readable and clear.
- 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 main weakness of this paper is the dataset used for the application experiments. The internal dataset is small (17 patients) and heterogenous (mix of pancreatic, head and neck, and “other” within these 17 patients) and was used exclusively for some analyses, while the external dataset was much larger (282 patients) containing only pancreatic cases. The heterogeneity of the local dataset poses some questions in how the OAR risk is assessed, as metric thresholds tends to vary considerably depending on the specific organ. Further, if posed as a generalized approach, rather than anatomical site-specific, more representation of non-pancreatic sites is important, particularly inclusion of prostate sites as this is the most common site of LDR brachytherapy procedures at most centres. 2.Related to the above weakness, another weakness of this paper is the choice of metrics for the ablation and comparative studies. The volumetric metrics presented are difficult to interpret in the context of the variable OAR-specific limits and D(2cc, 1cc, 0.5cc, 0.01cc) are more typical metrics to report for analysis of OAR protection.
- 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
In addition to the concerns outlined above, a few specific comments are included below:
Abstract 1.Please ensure that all acronyms are defined. 2.The source code link should also be in the main body of the work
Introduction 3.It should be specified explicitly early-on that this work pertains only to LDR brachytherapy applications to avoid ambiguity 4.As mentioned already above, I think it is important to clarify the scope of applications that this work will be tested on
Methods 5.The authors refer multiple times to generating “optimal plans”, however, it would be extremely beneficial to define what they are defining as optimal given the many factors and trade-offs involved in brachytherapy planning (CTV metrics, OAR protection, number of punctures, number of seeds, feasibility of needle trajectories)
Experiments & Discussion 6.It would be good to discuss limitations with regards to “difficult” cases such as those that include inaccessible sites based on patient anatomy (and potential intentional needle bending to achieve access) or cases where even the clinician-derived plan required compromised coverage. Were any of these included in the dataset? How might the proposed approach behave under these conditions? It would be helpful to include the clinical SR alongside the algorithmic metrics to help interpret the results. 7.Limitations of the work are not discussed.
- 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?
Overall, I think this is an interesting and well-done study with a strong level of methodological detail included. I believe this work has the potential to be impactful and of interest to the MICCAI community, particularly in the consideration of clinical decision-making factors; however, I do think the application scope should be clarified.
- 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 paper proposes SeedPro, a hierarchical reinforcement learning framework for brachytherapy seed placement that explicitly models trajectory-level and seed-level decisions. The method is clinically motivated because it jointly considers target coverage, OAR sparing, and procedural efficiency, especially puncture reduction. The reported results suggest that the framework can generate practically useful plans with improved OAR protection and fewer punctures than the main baseline.
- 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 problem is clinically relevant and the formulation is well aligned with how seed implantation is actually performed. 2) The hierarchical design is a sensible and technically interesting way to decompose a difficult combinatorial planning problem. 3) The paper evaluates not only dosimetric indices but also puncture number and seed number, which improves clinical relevance. 4) The method shows promising gains on both internal and external datasets, and the robustness study under different prescription settings is useful.
- 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 weaknesses are in validation rather than core idea. First, the internal dataset is relatively small, which makes it difficult to fully assess robustness across diverse anatomy and tumour sites. Second, the comparison study is mainly against one strong baseline; broader comparisons, especially against additional planning strategies or expert manual plans, would strengthen the claim of expert-level performance. Third, some aspects of the experimental setup and dataset construction could be clarified further, particularly the relationship between the internal and external cohorts and the practical availability of code/resources for reproduction.
- 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
This is a strong and clinically meaningful paper. The method is well motivated, and the emphasis on reducing puncture burden in addition to improving dosimetry is valuable. For the final version, I encourage the authors to improve experimental clarity, especially around external validation and reproducibility details, and to slightly tone down the expert-level claim unless supported by more direct clinician comparison.
- 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?
I recommend acceptance because the paper addresses an important clinical planning problem with an original and well-structured methodological solution. The hierarchical reinforcement framework is intuitive, the optimization objective is clinically grounded, and the reported quantitative gains are meaningful. While the empirical study could be broader, the strengths outweigh the limitations, and the paper is likely to be of interest to both the MIC and clinical planning communities.
- 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 proposes SeedPro, a hierarchical reinforcement learning–based framework for automated brachytherapy seed placement. The method constructs a multi-level trajectory-based state space to emulate clinical planning decisions and uses a coarse-to-fine RL strategy to optimize seed placement along selected trajectories. The objective function enforces target coverage (e.g., D90, V100) while minimizing OAR dose and procedural complexity (number of puncture needles). The approach is evaluated on a small in-house dataset and a larger external dataset, and compared against a prior method (BrachyPlan).
- 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.Automating brachytherapy planning, particularly reducing puncture needles while maintaining dosimetric quality, is highly relevant and impactful. 2.The hierarchical trajectory-based search combined with reinforcement learning is a reasonable and well-structured approach to handling the combinatorial nature of seed placement. 3.Explicitly optimizing the number of punctures (np) is valuable and aligns well with intraoperative considerations. 4.The inclusion of a large external dataset is a positive aspect and supports generalizability claims.
- 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 in-house dataset includes only 17 patients spanning multiple anatomical sites (e.g., pancreatic, head and neck, and others). Brachytherapy planning is highly site-specific, and this heterogeneity raises concerns about the robustness and clinical validity of the learned policy. 2.The inclusion of pancreatic cases is not well justified, as needle-based LDR seed implantation is not a widely adopted or standardized approach for pancreatic tumors. This raises questions about the realism and generalizability of the proposed framework. 3.The evaluation relies on comparison with a prior method (BrachyPlan) without direct benchmarking against expert clinical plans. As a result, claims of “expert-level” performance are not sufficiently substantiated. 4.The use of Vx metrics for OAR assessment limits interpretability and comparison with established clinical criteria (e.g., D2cc or Dmean). This makes it difficult to assess the true clinical benefit. 5.It is unclear whether key anatomical and procedural constraints (e.g., avoidance of critical structures, needle insertion feasibility) are fully incorporated, which is essential for real-world applicability. 6.While the framework prioritizes reducing the number of punctures, the number and spatial distribution of seeds per trajectory are not explicitly optimized. This may lead to suboptimal clustering or local dose heterogeneity, which are not analyzed.
- 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.
(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 addresses an important clinical problem and proposes a structured RL-based solution with promising potential. However, the current evidence is limited by a small and heterogeneous internal dataset, unclear clinical applicability for certain anatomical sites, and lack of direct comparison with expert clinical plans. Additionally, the use of non-standard evaluation metrics and absence of robustness analysis reduce confidence in the claimed improvements.
- 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
Q1: Dataset scope and design (R1, R2, R3): Although the internal dataset is much smaller than the external cohort, it was used for controlled sensitivity, ablation, and robustness studies. More importantly, SeedPro, developed and analyzed on this small-scale cohort, achieves superior performance on a substantially larger independent external dataset without any targeted tuning. This result directly demonstrates the robustness and reliability of the proposed framework.
Q2: Validation across additional anatomical sites (R1, R3): The site-specific clinical requirements are specified explicitly by the user rather than hard-coded into SeedPro. The framework takes as input CT images, CTV mask, OAR masks, and user-defined dosimetric objectives and constraints (by default following the OpenKBP standard). Under these predefined requirements, SeedPro solves the same optimization problem for all anatomical sites: generating the best plan that satisfies the specified dose constraints while minimizing invasiveness. Different treatment sites therefore differ only in the input masks and user-defined parameters, not in the underlying optimization mechanism. The current experiments are sufficient to support the claims and conclusions of this work. Validation on additional anatomical sites, including prostate, is beyond the scope of the present study and will be pursued as part of our future work.
Q3: Evaluation metrics (R1, R3): SeedPro is inherently compatible with standard clinical metrics. The reward function can directly optimize Vx, D2cc, D1cc, D0.5cc, Dmean, or any institution-specific criterion. We used Vx metrics in this study following consultation with collaborating clinicians, but this choice does not limit the framework. In practice, established metrics can be adopted without any architectural changes.
Q4: Comparison with expert plans (R2, R3): BrachyPlan was previously validated against expert manual plans and shown to outperform them (Medical Image Analysis, 2026, 107:103852). Since SeedPro consistently outperforms BrachyPlan, this provides indirect evidence that it exceeds expert-level planning quality.
Q5: Anatomical and procedural constraints (R3). SeedPro is able to automatically avoid obstacles during planning. These constraints are already incorporated in the Methodology section of our manuscript. Candidate trajectories that intersect user-specified OARs are automatically rejected, ensuring that generated plans avoid critical structures and satisfy essential procedural safety requirements.
Q6: Optimization of seed placement within each trajectory (R3). This is explicitly optimized by the low-level agent. The high-level agent selects needle trajectories, and the low-level agent determines the optimal positions of individual seeds along each trajectory. The reduced number of seeds required by SeedPro further confirms the effectiveness of this seed-level optimization.
Q7: Code and dataset availability (R2). The source code and all public data will be released upon acceptance. The anonymized repository link will be restored in the camera-ready version.
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.
The paper was evaluated by reviews with expertise in the field who identified limitations that lead to mixed enthusiasm for this work. Questions are raised about the design of the validation study and the application scope of the approach. However, the proposed methodology is of interest and addresses an important clinical problem. The strengths outweigh the weaknesses so I recommend acceptance.