List of Papers Browse by Subject Areas Author List
Abstract
External Radiation Therapy (ERT) is a key treatment in oncology, aiming to deliver high radiation doses to the Planned Target Volume (PTV) while minimizing exposure to surrounding healthy tissues and Organs At Risk (OARs). However, the proximity of PTVs to OARs, the presence of multiple OARs, and the time-consuming nature of manual subjective dose planning present significant challenges. While recent advancements in Deep Learning (DL) have led to various DL-based methods for dose prediction, it is still challenging to effectively capture multi-scale features and propagate essential information to related regions. In this work, we propose the Region-aware Attention Net (RANDose), which addresses these issues by integrating Multi-Scale Channel Spatial Attention (MSCSA), PTV Integration (PI), and Attention Fusion (AF) modules. Additionally, we introduce a Region-Aware Loss function to ensure accurate dose distribution within the PTV while minimizing radiation exposure to OARs. Experiments on the OpenKBP dataset demonstrate that RANDose outperforms existing models in both Dose Score and Dose Volume Histogram (DVH) Score, highlighting its superior performance.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/1645_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
N/A
Link to the Dataset(s)
N/A
BibTex
@InProceedings{ChoG._RANDose_MICCAI2025,
author = { Chowdary, G. Jignesh and Zhang, Tiezhi and Qian, Xin and Yin, Zhaozheng},
title = { { RANDose: A Region-aware Attention Network for Accurate Radiation Dose Prediction } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15974},
month = {September},
page = {530 -- 539}
}
Reviews
Review #1
- Please describe the contribution of the paper
To achieve accurate Radiation Dose Prediction, this paper proposes a Region-aware Attention Network (RANDose), which incorporates three key modules: Multi-Scale Channel Spatial Attention (MSCSA), PTV Integration (PI), and Attention Fusion (AF). Experimental results on a dataset demonstrate that the proposed method outperforms existing models in both Dose Score and Dose Volume Histogram (DVH) Score.
- 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 experimental pipeline is comprehensive, and the writing is clear and easy to follow. The focus on Radiation Dose Prediction carries clear clinical significance.
- 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.
- Although the proposed RANDose achieves state-of-the-art results on the OpenKBP dataset, most of its key modules—MSCSA, PI, and AF—are relatively straightforward combinations of existing techniques. While I do not oppose the use of such combinations for solving real-world problems, this approach provides limited methodological insights, making the paper less impactful from a methodological innovation standpoint and unlikely to meet MICCAI’s bar for methodological novelty.
- The proposed method is only validated on a single public dataset. Given the complexity of the network architecture, it remains unclear whether the model would generalize well to other Radiation Dose Prediction datasets.
- The code has not been released, which makes it difficult to verify the reproducibility of the results.
- 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.
- 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/2025/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 — could be rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
- The method lacks novelty and methodological insight.
- The experiments do not sufficiently demonstrate the generalizability of the proposed approach.
- The absence of code release limits reproducibility.
- 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 author address my concerns. I believe the paper could be accepted.
Review #2
- Please describe the contribution of the paper
The main contribution of the paper is to propose the region-aware attention net to capture multi-slice features and propogate essential information to related regions.
- 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.
Dose in the OAR remain unaffected desptie minor inaccuracies in PTV dose boundaries. *OAR: organ at risk, PTV: planning target volume
- 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.
DS and DVHS alone are insufficient to fully evaluate the model’s performance, and the reliance on segmentation data further limits the applicability of this approach.
- 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.
- 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/2025/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 — could be accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The trained model shows significant improvement of dose prediction in DS and DVHS. However, dependency on segmentation data inputs restricts the practicality of this approach.
- 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 appropriately address all of my concerns.
Review #3
- Please describe the contribution of the paper
a. Developed a deep learning-based dose prediction for head-and-neck cancer using an attention mechanism to inform planning target volume (PTV) while predicting dose distributions. b. Outperformed state-of-the-art deep learning-based dose prediction models.
- 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.
a. Performance improvements compared to the state-of-the-art dose prediction models on the same dataset. b. Good ablation studies on the model components of the proposed network modules and the loss function (Table 2).
- 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.
a. The loss function is effective but not novel, it was proposed in other studies with different names (e.g., masked L1 loss). b. Although this paper demonstrated the improved performance for overall dose scores for all regions (PTVs and OARs), no experiments explicitly showed the dose score (DS) and dose volume histogram score (DVHS) for the PTVs and OARs separately.
- 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.
- 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/2025/en/REVIEWER-GUIDELINES.html
a. In Fig. 2., the meaning of the centered dots was not clear. b. It is unclear whether the performance improvement was derived from the PI module or the L_PTV. The ablation studies should have shown clear contributions how the region-aware components collaborate, not separately. For example, the combination of Row ID 3 from Table 2(a) and Row ID 3 from Table 2(b) – so that we can validate which one contributes to ‘region-awareness’. c. Instead of incorporating the PTV maps via PI module, what will happen if the OAR information is incorporated together? Since the proposed model does not have causality of reinforcing the dose to the PTV, it is worth showing the effectiveness of OAR information to demonstrate the robustness of the proposed attention modules.
- 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 — could be accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
a. Although with the lack of thorough ablation studies about the region-awareness, the proposed model marginally outperformed the state-of-the-art models using the well-designed attention modules to incorporate PTV information.
- Reviewer confidence
Confident but not absolutely certain (3)
- [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.
Despite with lack of technical novelty and lack of validation on multiple datasets, this paper is well structured, and all the weaknesses are addressed in the rebuttal.
Author Feedback
We thank the reviewers (R) for their comments (C). We are encouraged that they found our work impactful in a clinical setting (R1, R2), appreciated the clear writing (R2), and were impressed by the results (R1, R3). Below, we address their questions:
R1C1 Performance metrics, and reliance on segmentation data DS and DVHS are the primary metrics used in prior SOTA methods. Based on your suggestion, we also evaluated our model using the Homogeneity Index (HI) [1], achieving a score of 0.782. As HI is not reported in existing works on the OpenKBP dataset, direct comparison is not possible. To address concerns about segmentation dependency, we tested a segmentation-free variant, which achieved DS:5.234, DVHS:3.952, and HI:2.846, showing a clear performance drop and underscoring the value of anatomical guidance through the OAR segmentation label. In future work, we plan to integrate automatic OAR segmentation to reduce reliance on manual labels while maintaining clinical relevance.
R2C1 Methodological insights While our model components may seem simple, each serves a distinct purpose. The PTV Integration (PI) module captures critical spatial information via multi-scale attention, enhancing dose prediction accuracy. The Region-Aware Loss aligns with clinical goals by balancing PTV coverage and OAR sparing. These custom-designed components are not off-the-shelf solutions but are tailored to address clinical challenges, contributing to consistent performance gains. We value methodological innovation and believe RANDose offers meaningful, task-specific contributions relevant to the MICCAI community.
R2C2 Single-Dataset Evaluation. Currently, OpenKBP is the only publicly available dataset for radiation dose prediction.
[R2][C3] Code Availability As noted in Section 3.2, we will release the code.
R3C1 Loss Function. Our Region-Aware Loss extends the standard masked L1 loss, which applies L1 only to the region of interest (PTV). In contrast, our loss includes three components: L1 over all voxels, within the PTV, and over the OARs. As suggested, we ran an ablation using only the standard masked L1 loss, which resulted in DF: 2.364 and DVHS: 1.443. This drop in performance shows the value of including loss terms beyond the PTV.
R3C2 Evaluation in PTV and OAR Regions. Following the suggestion, we computed the DS, and DVHS separately for the PTV and OAR regions. The results are shown below: |Region| DS |DVHS| |PTV |1.873|1.004 | |OAR |2.507|1.316 | These results show that the model performs more accurately on the PTV region, which is expected as the architecture and loss are explicitly designed to prioritize target coverage while minimizing dose to the OARs.
R3C3 Dots in Fig. 2. The centered dots in Figure 2 represent multiple layers or blocks.
R3C4 Other Ablations Based on the suggestion, we conducted an ablation study with a fixed MSCSA (M) module to evaluate the PI module and Region-Aware Loss (R) individually and together. |M|R|PI| DS |DVHS| |Y|Y| |2.993| 2.353 | |Y| | Y|2.843| 2.198 | |Y|Y| Y|2.589| 1.621 | These results show that while both the R and the PI modules contribute individually to performance improvements, their combination results in the most significant gains across all evaluation metrics.
R3C5 Effect of using OAR Instead of PTV via PI Module We ran two additional experiments to assess the impact of incorporating OARs into the PI module. Replacing PTVs with OARs led to worse performance (DS:2.346, DVHS:1.543), while using both offered slight improvements but still underperformed compared to using only PTVs (DS:2.256, DVHS:1.415). These results highlight the importance of focusing on PTVs for effective dose modeling. That said, exploring dual-region or adaptive guidance remains a promising direction for future work.
References [1] Fu, L. et al. MD-Dose: A diffusion model based on the Mamba for radiation dose prediction. In 2024 BIBM.
Meta-Review
Meta-review #1
- Your recommendation
Invite for Rebuttal
- If your recommendation is “Provisional Reject”, then summarize the factors that went into this decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. You do not need to provide a justification for a recommendation of “Provisional Accept” or “Invite for Rebuttal”.
N/A
- 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. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
N/A
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. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
All reviewers agree that the concerns have been addressed by authors in the rebuttal (although technical novelty is relative minor). Recommend acceptance.
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. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
This paper received three positive final ratings from the reviewers. The AC concurs with the reviewers’ evaluations, and this paper is good to be accepted.