List of Papers Browse by Subject Areas Author List
Abstract
Adaptive radiotherapy (ART) improves treatment precision by adapting to anatomical changes, but its clinical adoption is limited by high costs, patient burden, and institutional variability. To address this, we propose a robust multi-omics nomogram for predicting ART eligibility in nasopharyngeal carcinoma (NPC) patients by integrating multi-modality Genomap signatures with clinical factors. Using retrospective data from 311 patients at Queen Elizabeth Hospital (training set) and 192 patients at Queen Mary Hospital (external test set), we extracted 7,956 radiomics features from six regions-of-interest (ROIs) across contrast-enhanced computed tomography (CECT), magnetic resonance imaging (MRI), and dose modalities, alongside 132 geometric features capturing spatial relationships between ROIs. Feature selection via LASSO identified 35 radiomic, 8 dosiomic, and 4 geometric features for analysis. The Genomap model achieved an accuracy of 80% and an AUC of 90% across modalities, while the integrated nomogram demonstrated superior performance with 88% accuracy and 96% AUC. Our results show that Genomap ensures generalizability and robustness, providing a reliable tool for personalized ART planning in NPC patients.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/2540_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{SheJia_Boosting_MICCAI2025,
author = { Sheng, Jiabao and Li, Zhe and Zhang, Jiang and Lam, Saikit and Chen, Zhi and Xing, Lei and Cai, Jing},
title = { { Boosting Generalizability in NPC ART Prediction via Multi-Omics Feature Mapping } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15974},
month = {September},
page = {34 -- 43}
}
Reviews
Review #1
- Please describe the contribution of the paper
The paper focuses on Nasopharyngeal Carcinoma (NPC), a cancer with high prevalence in East and Southeast Asia. NPC is typically treated with radiation therapy whose planning does not account for anatomical changes during treatment (and it can reduce the accuracy of therapy). The authors focus on Adaptive Radiotherapy (ART), as a more precise approach that updates treatment plans in real time but whose usage remains limited due to high costs, patient burden, and technical variability across institutions.
Specifically, they propose a multi-omics-based predictive model to determine which NPC patients are likely to benefit from ART. Multi-omics refers to the integration of multiple layers of biological or biomedical data to capture a more comprehensive view of the patient condition. In medical imaging, multi-omics includes combining image-derived features (radiomics), dose distribution data (dosiomics), anatomical contour information (contouromics), and clinical metadata.
The proposed scheme is articulated into two main steps. Firstly, they construct a Genomap, a map that transforms heterogeneous multi-omics data (from CT, MRI, dose maps, and contours) into structured 2D feature maps. Secondly, a CNN recalibrates these transformed features and builds a predictive nomogram (a visual tool used to predict the probability of the outcome, as a user-friendly chart for doctors).
- 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 demonstrates that the proposed workflow is highly generalizable, thus usable across diverse clinical settings. The experimental part, in fact, is based on patients treated with radiotherapy from Queen Elizabeth Hospital for model training and from Queen Mary Hospital for external validation. This methodology is highly consistent with the usage of the feature maps, reducing variability across institutions.
- 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 suffers from serious writing issues that significantly hinder readability and comprehension. Acronyms are overused throughout the text, often without proper definition, making it difficult for readers to follow. Furthermore, many technical terms are introduced without adequate explanation, which restricts the accessibility of the paper to a very specialized audience. In addition, the manuscript contains numerous language errors, including typos, incorrect punctuation, and inconsistent spacing, all of which contribute to an overall lack of clarity and polish.
An additional weakness of the paper is the lack of a sufficiently rich bibliography, which results in poor contextualization of the proposal in the state-of-the-art techniques. As a consequence, it is difficult to clearly understand how the proposed method builds upon or differs from existing approaches precisely. In particular, the specific contribution of this work relative to references [13] and [17] remains unclear and should be more explicitly explained.
- 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.
- 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
Given the complexity of the workflow, a more consistent and coherent labeling of the steps throughout the paper would greatly assist the reader. For example, Figure 1 uses different terminology and is incomplete, lacking the right-hand portion. In Section 2.1, you describe only the left-hand side of Figure 1, which refers to the preprocessing step. If I understand correctly, this preprocessing is not part of your main contribution (in that case, the detailed illustration in Figure 1 might be unnecessary). Additionally, there’s an inconsistency regarding the number of ROIs: are you using 8 ROIs as shown in the figure, or 6 as stated in Section 3.1?
The relationship between the notation in Section 2.3 and that in Section 2.2 is also unclear. The equation in Section 2.2 appears to use \bar{x} to denote mean values.
In Section 3, the results are only briefly discussed and lack a clear and structured description. In Section 3.2, the methods used for comparison are not sufficiently explained. Figure 3.A, in particular, should be explicitly referenced and interpreted in the text, as it is not straightforward to understand what it represents. The proposed representation intended to aid clinical decision-making does not come across as intuitive to a more general audiance.
- 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?
While the proposed method and the clinical application addressed are certainly of interest to the MICCAI community, the paper does not meet the overall standards typically expected at this conference. Issues in writing quality, clarity of presentation, and lack of methodological detail significantly limit its impact and accessibility.
- Reviewer confidence
Not confident (1)
- [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.
In the rebuttal, the authors promise to significantly revise the submission by adding new tables and, more importantly, entirely new subsections to improve the paper. While I appreciate their willingness to address the weaknesses, such extensive modifications go beyond what is typically acceptable in a rebuttal phase.
Review #2
- Please describe the contribution of the paper
The paper presents an innovative approach by integrating multi-omics features through the Genomap framework to predict ART eligibility in NPC patients. Here are the main contributions:
-
Introduces a novel Genomap framework that integrates multi-omics data (radiomics, dosiomics, and geometric features) into a unified feature map.
-
Leverages a normalized partial correlation coefficient matrix and Gromov-Wasserstein optimal transport for robust multi-modal feature mapping.
-
- 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.Novel Genomap framework for integrating multi-omics data into a unified feature map.
-
Innovative use of a normalized partial correlation coefficient matrix and Gromov-Wasserstein optimal transport for robust feature mapping.
-
Comprehensive integration of multiple modalities (radiomics, dosiomics, and geometric features) with clinical factors.
-
- 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.
While the proposed model demonstrates promising performance and improved generalizability across data from two institutions, several limitations need to be addressed:
-
There is a risk of overfitting. The author did not clearly mention the train-valid-test split, how did the author split multisource data (there were only 503 cases in total and this scale for a CNN-based regression is relatively small)? More rigorous cross-validation procedures and regularization strategies should be considered to ensure that the model’s performance is robust and not merely a result of overfitting to the specific datasets used in this study.
-
The use of the Gromov-Wasserstein discrepancy for optimal feature mapping, while interesting, typically involves significant computational overhead. The paper does not provide sufficient details regarding the runtime or scalability of the approach.
-
It would be beneficial to see a discussion on how the model could be integrated into clinical workflows and evaluated in a prospective manner.
-
Abbreviations need to be clarified when they were firstly used, e.g., GTVn.
-
- 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.
(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?
Although there are some issues, this study is interesting and has clinical values. The author needs to carefully address the overfitting concerns and clarify their training strategies.
- 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 adapts Genomap to radiotherapy planning by mapping multi-omics features (CECT, MRI, dose, geometry) into a structured 2D space using normalized partial correlation and Gromov-Wasserstein transport. This enables a CNN + logistic regression model to predict ART eligibility in NPC patients. The main contribution is improving cross-institution generalizability, with strong results on a two-center dataset (AUC 0.95/0.96), outperforming standard machine learning baselines.
- 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 introduces a structured feature mapping approach using normalized partial correlation and Gromov-Wasserstein transport, enabling geometry-aware integration of radiomic, dosiomic, and geometric features—this is a novel adaptation in the context of medical imaging. (2) The method demonstrates strong cross-institution generalizability, with consistently high AUCs on both internal (QEH) and external (QMH) datasets, which is a critical challenge in clinical model deployment.
- 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 core mapping method (Genomap) is not novel and is directly adapted from prior work in single-cell data analysis [17]; the paper does not introduce algorithmic innovations beyond applying it to medical imaging. (2) The CNN architecture is under-specified, with no details on network depth, filter sizes, or training setup. This limits reproducibility and weakens the technical rigor of the method section. (3) The paper lacks statistical significance testing (e.g., confidence intervals, p-values) to support claims of superiority over baselines, and does not include ablation studies isolating the effect of Genomap mapping.
- 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 provide sufficient information for 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.
(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?
I recommend a weak accept mainly due to the paper’s strong application value and solid empirical results. The adaptation of Genomap to multi-omics ART prediction is novel in this context and effectively addresses cross-institution generalizability, which is a known limitation in existing models. The external validation is well-executed and shows consistent performance. However, the technical contribution is more application-driven than methodological, and the lack of architectural details and statistical significance testing weakens the overall rigor. With additional methodological clarification, this work would be stronger.
- 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.
This paper proposes a robust multi-omics prediction pipeline that integrates Genomap feature mapping with clinical data to improve adaptive radiotherapy (ART) eligibility predictions for NPC patients. The approach is innovative in how it addresses cross-institutional variability, validated on a large external dataset, and demonstrates substantial improvements in predictive accuracy and generalizability.
Author Feedback
R#2 // Q1: In Section 3.1, we report training on QEH data and testing on QMH data. This cross-center validation, as commonly used in clinical practice, more realistically assesses the model’s robustness to inter-domain distribution differences (e.g., imaging protocols, device vendors, and patient populations) compared to cross-validation.
Q2: These are interesting questions but beyond the current scope. We plan to explore them in future work.
Q3: Our method can be integrated into the clinical workflow between radiotherapy planning and treatment initiation as a decision-support tool to help physicians assess dose appropriateness. For example, it may identify risks of normal tissue damage from overdosing, enabling informed adjustments. We will add a new Section 3.5 to further discuss this point.
Q4: We will include a notation table in the revised manuscript to explain all abbreviations used in our paper.
R#3 // Major Weaknesses: We will include a notation table explaining all abbreviations and define technical terms (e.g., Gromov-Wasserstein, Nomogram) at first appearance. The manuscript will be thoroughly proofread to fix typos, punctuation errors, and spacing inconsistencies. While the main contributions are outlined in the introduction, we will add references as needed to better clarify how our work builds upon and differs from previous studies.
Optional Comments: We will correct the screenshot error in Fig. 1. While the preprocessing workflow provides essential background, we will consider simplifying it. Features were initially extracted from 8 ROIs, but only 6 were retained after selection. This will be clarified in the caption of Fig.1. We will double-check notation in Sections 2.2 and 2.3 and revise \bar{x} to \mu. A new Section 3.5 will be added to organize results by task and metric, compare input strategies or models, and discuss findings in context. To improve contextualization, we will add references that were previously omitted due to space limits. Fig. 3A’s caption will also be updated to explain interpretation, including risk factors and total score computation.
R#4 // Q1: Rather than directly adapting prior work, we extend its applicability to a clinically meaningful task using new data modalities and evaluation settings. Genomap uncovered associations among GTVnp, GTVn, contralateral/ipsilateral parotid gland, brainstem, and spinal cord across multi-omics data. Features from different institutions were mapped into a unified space to reduce inter-institutional variability. Genomap-generated feature weights were output separately for each omics type (MRI, CECT, dose, and Contouromics) and integrated with clinical features to construct a multi-omics nomogram for assessing whether patients require replanning based on current radiotherapy plans.
Q2: We will add a subsection in Section 3 to clarify implementation details: the model architecture consists of 10 layers, including an input layer, one convolutional layer (kernel size = 3, channels = 8), two ReLU layers, three fully connected layers, one dropout layer, a softmax layer, and a final classification layer. The model was optimized using the Adam optimizer with weight decay (learning rate = 0.001, decay = 0.00001).
Q3: We adopted a cross-center validation strategy, widely used in clinical practice, i.e., training on one institution’s data and testing on another’s. This better reflects real-world scenarios by evaluating generalizability under distribution shifts. We acknowledge that statistical significance testing adds robustness and plan to include confidence interval estimation or permutation testing in future work.
In Figs. 2A and 2B, we compare ablation results of models with and without Genomap-based transformation (e.g., Random Forest, XGBoost, Logistic Regression, Ridge Regression). Fig. 2B shows that ‘data from different institutions can still achieve reliable results after applying Genomap,’ validating its generalizability and robustness.
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”.
The paper has potential, but it must improve in its presentation and writing.
- 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.
Reject
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
The paper proposes a novel approach taken from genomics data analysis to integrate a variety of image based features through a partial correlation matrix with Gromov Wasserstein distance computation to represent features through their relatedness and then calibrate them via a CNN to predict patients likely to benefit from adaptive radiotherapy. In this respect, the paper is interesting. However, several key details are missing in the methods - including how the data was trained and evaluated, addressing class imbalance, construction of CNN for calibration, etc. Details of how the features were normalized to compute the partial correlation matrix is also missing. It’s unclear if the feature preselection using LASSO was performed using all the data or a subset. Paper also provided names of institutions including the home institution, violating the confidentiality requirements of the review process. It is also not possible to perform major revisions of manuscript including adding tables as part of the revision. For these reasons, the paper cannot be accepted in its current form.
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’
The authors have addressed main issues raised by reviewers. Please revise the final version according to reviewers’ suggestions.