List of Papers Browse by Subject Areas Author List
Abstract
Kidney tumors can be highly heterogeneous from the microscopic to the macroscopic scale. To address this, we propose a sparsity-informed probabilistic integration of radiomics and pathomics for kidney cancer analysis. We construct radiology and pathology graphs to model spatial correlations, then use a probabilistic method and graph neural networks to identify biomarkers and aggregate spatial features. Our validation shows that this integrated approach significantly outperforms traditional methods in kidney survival analysis, achieving a notable improvement of 0.084 in the concordance index, enabling better prognostic assessments for kidney cancer patients. The source code has been released by https://github.com/shangqigao/RadioPath.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/1823_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/shangqigao/RadioPath
Link to the Dataset(s)
TCGA-KIRC: https://www.cancerimagingarchive.net/collection/tcga-kirc/
TCGA-KIRP: https://www.cancerimagingarchive.net/collection/tcga-kirp/
TCGA-KICH: https://www.cancerimagingarchive.net/collection/tcga-kich/
TCGA WSIs: https://portal.gdc.cancer.gov
BibTex
@InProceedings{GaoSha_Probabilistic_MICCAI2025,
author = { Gao, Shangqi and Gao, Shangde and Machado, Ines and Crispin-Ortuzar, Mireia},
title = { { Probabilistic Integration of Renal Cancer Radiology and Pathology Using Graph Neural Networks } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15971},
month = {September},
page = {563 -- 572}
}
Reviews
Review #1
- Please describe the contribution of the paper
Authors present a probabilistic model to integrate radiology and pathology sources of information. In order to learn spatial sparsity importance, the proposed model uses a Student’s prior and graph neural networks. Under the clinical application of predicting overall survival in kindey cancer, this model was assessed using highly heterogeneous public 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 model proposal, which demonstrate how the proposed method, SPARRA, outperformed other aggregation methods.
- Significant amount of data to provide promising results (205 subject cohort including for each one WSI, CT scan, and clinical record)
- Performance evaluation is based on testing different feature extractors, aggregation models, survival models, and finally evaluated using different metrics to provide an extensive assessment.
- 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.
- Data description and data splitting (testing purposes) for experiments need clarifications
- This study does not evaluate how the differences in pathologies/diseases across the datasets (e.g. Kidney Renal Clear Cell Carcinoma or Cervical Kidney renal papillary cell carcinoma) could affect results, and these pathologies were distributed in data splitting.
- 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
Data description:
- In Dataset description, correct the name of the third public dataset: CGA-KICH
- Not clear from the 205 subject included in the study, how many come for each of the 3 used public datasets: TCGA-KIRC [18], TCGA-KIRP [19], TCGA-KIRH [20].
- How do the extracted features differ depending on the specific pathology or cohort (TCGA-KIRC→ Kidney Renal Clear Cell Carcinoma, TCGA-KIRP→ Cervical Kidney renal papillary cell carcinoma, TCGA-KICH → Kidney Chromophobe Collection).
- For the last experiment what is the survival threshold and how many samples are per risk-group.
Feature selection:
- How many features (radiomics and pathomics) does the study have before constructing the radiological and pathological graphs? After the univariate feature selection, how many features per utilized tool selected or just some of them (in such a case which ones)?
- Why do not consider combine the most relevant features of each feature extractor?
Survival analysis:
- Clarify why according to data description testing set is composed of 41 subjects, but results on testing are reported with 19 individuals.
- In Fig. 3, I recommend to plot each metric value as a bar along with the standard deviation, since in the current manner (line plot) looks like each result is connected in some way to the previous or the following ones
- 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?
This is an interesting work, however some issues more related to data than with the SPARRA model, should be addressed.
- 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 #2
- Please describe the contribution of the paper
The main contribution of the paper is the development of a sparsity-informed probabilistic integration of radiomics and pathomics for kidney cancer analysis. The authors propose a novel approach that constructs radiology and pathology graphs to model spatial correlations, and then use probabilistic methods combined with graph neural networks to identify biomarkers and aggregate spatial features. The experimental validation demonstrates that this integrated approach significantly outperforms traditional methods in kidney survival analysis, with a notable improvement of 0.084 in the concordance index, offering enhanced prognostic assessments for kidney cancer patients.
- 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.
-Enhanced Spatial Sensitivity: By incorporating a Student’s t priorand utilizing graph neural networks, the method improves spatial sparsity handling, enhancing the model’s ability to learn and prioritize spatial features. -Rigorous Mathematical Derivation: The paper includes detailed mathematical derivations that provide a solid theoretical foundation for the proposed model, ensuring a robust understanding of the methodology.
- 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 of the paper are related to the complexity and clarity of the mathematical derivations, especially in the section on Variational Inference. The mathematical formulas and reasoning behind them are difficult to understand, and the paper lacks sufficient explanation of the motivation for the formulas and the dimensions of the variables. A more detailed description of these aspects would help clarify the method and make the approach more accessible and comprehensive.
- 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
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?
My recommendation is based on the innovative approach presented in the paper, particularly the mathematical derivation of the proposed method. While the idea of integrating radiomics and pathomics through a probabilistic model is promising, the mathematical formulation, especially in the Variational Inference section, is difficult to understand. The paper would benefit from more detailed explanations of the motivation behind the formulas and clearer descriptions of the variables’ dimensions. Given that multimodal fusion has already been extensively explored, the mathematical complexity of the approach makes it harder to fully appreciate its contribution without more accessible explanations. This led to my overall score.
- 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
Kidney cancer shows significant heterogeneity across genomic, histologic, and radiomic levels. This work proposes a unified framework to integrate these diverse sources for improved analysis.
- 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 framework for kidney cancer survival analysis by integrating radiomics and pathomics through a sparsity-informed probabilistic model. It effectively addresses spatial aggregation challenges using radiology and pathology graphs, incorporates a Student’s t prior to enhance spatial sparsity, and applies GNNs to learn spatial importance. The method is thoroughly evaluated across radiomics, pathomics, and their integration, demonstrating its effectiveness.
- 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.
Overall, the paper offers a strong contribution; however, the conclusion section is notably weak and should include a discussion of the current work’s limitations.
- 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
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 nature of the problem addressed by this research, along with the proposed framework, has strong potential to impact survival prediction in kidney cancer.
- 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
All reviewers acknowledged the innovative approach and thorough evaluations presented in this work. We sincerely appreciate their insightful comments and constructive suggestions, which we have carefully addressed as detailed below. Our source code is ready for release at GitHub.
Q1: How many subjects for each subtype? (Reviewer 1) A1: Among 205 subjects, we have 180 subjects from TCGA-KIRC, 15 patients from TCGA-KIRP, 10 patients from TCGA-KICH. We will clarify this in our revision.
Q2: Feature difference between subtypes. (Reviewer 1) A2: For pathomic features, previous foundation models have shown that extracted features are very different between subtypes, with a very high subtyping precision [8].
Q3: Survival threshold and the number of subjects for each risk group.(Reviewer 1) A3: We separated the low- and high-risk samples based on their risk scores by comparing them to the mean risk score, which is a widely used strategy in previous work [9]. By cross-validation, we have 94 subjects identified as high-risk, while 111 patients identified as low-risk for our method.
Q4: How many features before graph construction, and how many selected by the univariate feature selection? (Reviewer 1) A4: Before graph construction, the feature dimension of radiomics is 768 for the SegVol feature extractor, while that of pathomics is 1024 for the UNI feature extractor. On average, we have 626,400 radiomic feature vectors per CT scan, while 15,544 pathomic feature vectors per WSI. After graph construction, we have 15,660 radiomic feature vectors per CT scan and 2,364 pathomic feature vectors per WSI for spatial aggregation. After the spatial aggregation and univariate feature selection, the feature dimension ranges between 100 and 500, depending on specific feature aggregation methods.
Q5: Why not consider combine the most relevant features of each feature extractor? (Reviewer 1) A5: The radiomic feature vectors and pathomic feature vectors have different dimensions and are in different spaces. Besides, different feature extractors have highly heterogeneous neural network architectures. These differences make it difficult to directly measure the relevance between radiomic and pathomic features.
Q6: Clarification on survival analysis. (Reviewer 1) A6: We would like to clarify we did 19 studies. For each study, we evaluated models on 41 testing subjects based on the data splitting for cross-validation.
Q7: Limitations should be discussed. (Reviewer 2) A7: This work has not studied disease-free survival or subtype-specific analyses. We will discuss this in our conclusion.
Q8: Clarification of mathematical derivations. (Reviewer 3) A8: The variational inference is a well-known method to solve intractable MAP problems based mean-field theory in statistics. We will revise the descriptions of that part to avoid any confusions.
Meta-Review
Meta-review #1
- Your recommendation
Provisional Accept
- 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