List of Papers Browse by Subject Areas Author List
Abstract
Esophageal varices (EV), a serious health concern resulting from portal hypertension, are traditionally diagnosed through invasive endoscopic procedures. Despite non-contrast computed tomography (NC-CT) imaging being a less expensive and non-invasive imaging modality, it has yet to gain full acceptance as a primary clinical diagnostic tool for EV evaluation. To overcome existing diagnostic challenges, we present the Multi-Organ-cOhesion-Network (MOON), a novel framework enhancing the analysis of critical organ features in NC-CT scans for effective assessment of EV. Drawing inspiration from the thorough assessment practices of radiologists, MOON establishes a cohesive multi-organ analysis model that unifies the imaging features of the related organs of EV, namely esophagus, liver, and spleen. This integration significantly increases the diagnostic accuracy for EV. We have compiled an extensive NC-CT dataset of 1,255 patients diagnosed with EV, spanning three grades of severity. Each case is corroborated by endoscopic diagnostic results. The efficacy of MOON has been substantiated through a validation process involving multi-fold cross-validation on 1,010 cases and an independent test on 245 cases, exhibiting superior diagnostic performance compared to methods focusing solely on the esophagus (for classifying severe grade: AUC of 0.864 versus 0.803, and for moderate to severe grades: AUC of 0.832 versus 0.793). To our knowledge, MOON is the first work to incorporate a synchronized multi-organ NC-CT analysis for EV assessment, providing a more acceptable and minimally invasive alternative for patients compared to traditional endoscopy.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/0915_paper.pdf
SharedIt Link: pending
SpringerLink (DOI): pending
Supplementary Material: https://papers.miccai.org/miccai-2024/supp/0915_supp.pdf
Link to the Code Repository
N/A
Link to the Dataset(s)
N/A
BibTex
@InProceedings{Li_Improved_MICCAI2024,
author = { Li, Chunli and Zhang, Xiaoming and Gao, Yuan and Yin, Xiaoli and Lu, Le and Zhang, Ling and Yan, Ke and Shi, Yu},
title = { { Improved Esophageal Varices Assessment from Non-Contrast CT Scans } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
year = {2024},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15005},
month = {October},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper discussed a critical problem of accessing EV from NC-CT. This is a challenging image analysis task of extracting multi organ features for portal hypertension. The authors propose MOON, a holistic approach to assist diagnose EV. This work also includes a NC-CT dataset comprised of 1255 patients, with three grades of severity.
- Please list the main strengths of the paper; you should write about 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.
- This paper worked on a critical problem on accessing EVm, the background and clinical application has potential contribution to the community.
- The experiment design is well-structured, proved effectiveness ot the method.
- Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.
This paper lacks basic discussion of techniques, the introduction has many decisive statements, but did not cite or discuss rationales, which make the proposed method description subjective. What is the challenges and background of current literature EV access using NC-CT? What’s the current adopted methods, baselines or technical backs? The method used UniFormer as backbone, takes multi-organ features as input, also incorporates a organ detection module. The authors can highlights the innovation of the method. Also, for portal hypertension assessments, only liver and spleen features are required? Is there any citation or proof to support the metric.
- 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.
- Do you have any additional comments regarding the paper’s reproducibility?
The paper did not mentioned the accesbility of code and project.
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html
The authors can discuss more on the technical merits of methods.
- 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
Weak Accept — could be accepted, dependent on rebuttal (4)
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Overall, the paper is well structured, the problem and solution has potential value to the community though it’s a bit lack of technical merits.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
N/A
- [Post rebuttal] Please justify your decision
N/A
Review #2
- Please describe the contribution of the paper
The paper describes a system to classify the severity of esophageal varices from non-contrast-enhanced chest CT. A deep-learning based pipeline called the Multi-Organ Cohesion Network (MOON) is proposed. First, an nnU-net is used to localize the spleen, liver, and esophagus. These regions are cropped and fed into parallel pipelines which encode the images using the Uniformer, sharing information between organs at each layer using the Organ Representation Interaction modules. On the esophageal path, the Hierarchical Feature Enhancement module integrates features at multiple scales. Results from each organ are fused and fed through a classification unit. The model is trained to optimize an ordinal classification loss, regularized by Canonical Correlation Analysis loss, which penalizes incoherent results from the three parallel pipelines. The results show better classification accuracy than state-of-the-art results using radiomics on dynamic contrast CT.
- Please list the main strengths of the paper; you should write about 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 use of clinical knowledge to inspire the network architecture and inputs (namely the use of not just the esophagus, but also the liver and spleen due to their involvement in the condition) is insightful.
The training data set and external validation set are both relatively large, which strengthens the validation.
By relying on non-contrast CT, the system requires a minimal radiation dose compared to others which require contrast.
The ablation study and tests of single-organ pipeline performance show that the various architectural choices and innovations (the Organ Representation Interaction and Hierarchical Feature Enhancement units are in fact providing a benefit.
- Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.
The main weakness I see is that, unless I am reading the figures wrong, it appears that the ablation study (Fig 2, Supp Fig 1) shows that removing the HFE actually has a larger effect than ORI or CCA, while the main text of the results states that ORI was the most significant. I don’t understand where this conclusion is coming from.
Furthermore, while the ablation study and single organ analysis demonstrate the value of the addition of ORI and HFE, one comparison that is missing, in my opinion, is that of the esophagus alone, with HFE. Given that the removal of HFE has the largest effect on performance (based on Fig 2, and supp Fig 1), I wonder where the esophagus pipeline on its own, with HFE, would fall? If I am misunderstanding, and HFE was included in the single-organ results for esophagus, then it would help to clarify that in the manuscript.
Finally, the description of the dataset is limited: while it is stated that the test set was obtained from a single institution, no mention of number of institutions is given for the larger training set. If the training set was also from one institution, were both datasets from the same institution? It would also be helpful to know how variable the acquisition protocol was – are these low-dose images? is iterative reconstruction used? etc.
- Please rate the clarity and organization of this paper
Very 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.
- Do you have any additional comments regarding the paper’s reproducibility?
A few things should be clarified to ensure reproducibility:
-
are the segmentations applied only to crop the images into each pipeline, or are they also used to mask out the background?
-
the fusion layer is not actually described in the methods - it is only briefly explained that several methods were used to fuse the results from each pipeline in results paragraph on page 7. This makes that paragraph harder to understand, as well, when it refers to “post-fusion strategies”
-
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html
One thing that would be helpful is a clearer specification of the contributions of the work. While it is made clear that the creation of a system with better accuracy is the major contribution, I also wonder about things like the ORI and HFE. By the description, these appear to be novel, and provide a substantial benefit to the system. If this is right, that novelty could be highlighted in the introduction of the system.
Beyond that, the main constructive feedback i have for the authors is in my answers to 6 and 9. To reiterate:
I think it would help to clarify how the conclusion that ORI was most significant was reached, given the results in Fig 2 and Supp Fig 1.
Speaking of Fig 2, the 4 groups of bars are labelled AUC (>=G2), AUC (G3), AUC (>= G2), AUC (G3). It is not stated what the difference between the first and third, or second and fourth are, and they are labelled identically. I assume the answer is the same as for Supp fig 1, which shows that the first two are for the “concat” fusion strategy, and the second two are for the FiLM strategy, but this should be shown in Fig 2 itself as well.
I think the authors should briefly describe the use of multiple fusion layer implementations in the methods section.
I think the authors should specify how the segmentations are used more clearly: is the background masked out? Does the cropping include some padding, or just the bounding box of the segmentation? etc
I also think it would help to give a bit more information about the imaging - how many institutions, what protocols, etc.
Finally, a minor point: there are several places where sentences start with no space after the period ending the previous sentence.
- 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
Weak Accept — could be accepted, dependent on rebuttal (4)
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The paper provides a valuable contribution solving an important clinical problem. The use of non-contrast CT, which requires lower radiation dose is important. The validation of the system looks good. Though there are a few clarifications and weaknesses I discuss in my review, I think the paper should be accepted.
The only reason I am giving a 4 rather than a 5 is the concern about the apparent discrepancy between Fig 2 and Supp Fig 1 showing the greatest performance drop when HFE is dropped, while the main text concludes that ORI was most significant. I think this must be clarified before the paper is accepted, hence I am making my recommendation dependent on rebuttal.
- Reviewer confidence
Somewhat confident (2)
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
N/A
- [Post rebuttal] Please justify your decision
N/A
Review #3
- Please describe the contribution of the paper
-MOON framework which provides a cohesive assessments unifying imaging features from different organ sites useful in diagnosis of Esophageal Varices. -The Organ Representation interaction and the Hierarchical Feature Enhancement modules - allowing to identify the important features at different scales from multiple organs.
- Please list the main strengths of the paper; you should write about 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.
- Novel framework (MOON) for multi-organ based diagnostic assessment from a NC-CT scan.
- Useful ORI and HFE modules combining features from multiple organs and different scales.
- Superior metrics for classification on Independent Test set with model using MOON compared to only Eso. based model.
- Comprehensive ablation study showcasing comparative qualitative results, validating the quantitative metrics.
- Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.
- Need to show the improvement in AUC is statistically significant. Use Delong’s Test to compare classification metrics.
- Please rate the clarity and organization of this paper
Excellent
- 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.
- Do you have any additional comments regarding the paper’s reproducibility?
No
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html
Good paper. I liked the ablation study identifying ORI module being more critical than the HFE module or the CCA loss .
- 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
Accept — should be accepted, independent of rebuttal (5)
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Novel method for multi organ based diagnostic assessment is developed. The importance of the novel method is validated by strong quantitative metrics and useful qualitative comparison. Comprehensive ablation studies done for understanding the findings.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
N/A
- [Post rebuttal] Please justify your decision
N/A
Author Feedback
We sincerely thank the area chairs and anonymous reviewers for their invaluable time and efforts spent on our work. We are deeply grateful for their constructive feedback, and will diligently take all the feedback into account to improve the quality of this work.
Meta-Review
Meta-review not available, early accepted paper.