List of Papers Browse by Subject Areas Author List
Abstract
Assessing lymph node (LN) metastasis in CT is critical for esophageal cancer treatment planning. While clinical criteria are commonly used, the diagnostic accuracy is low with sensitivities ranging from 39.7% to 67.2% in previous studies. Deep learning would have the potential to improve it by learning from large-scale accurately labeled data. However, from the surgical procedure in LN dissection, pathological report only indicates the number of dissected LNs in each lymph node station (LN-station) with the number of metastatic ones found in the respective LN-station. So, it is difficult to establish one-to-one pairing between LN instances observed in CT and their metastasis status confirmed in the pathological report. In contrast, gold reference labels on LN-station metastasis can be readily retrieved from pathology reports at scale. Hence, instead of distinguishing LN instance metastasis, we directly classify LN-station metastasis using pathology-confirmed station labels. We first segment mediastinal LN-stations automatically to serve as input for classification. Then, to improve classification performance, we automatically segment all visible LN instances in CT and design a new LN prior-guided attention loss to explicitly regularize the network to focus on regions of suspicious LN. Furthermore, considering the varying appearances and contexts of different LN-station, we propose a station-aware mixture-of-experts module, where the expert is trained to specialize in a group of LN-stations by learning to route each LN-station group tokens to the corresponding expert. We conduct five-fold cross-validation on 1,153 esophageal cancer patients with CT and pathology reports (the largest study to date), and our method significantly outperforms state-of-the-art approaches by 2.26% in AUROC.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/1419_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{LiHao_Metastatic_MICCAI2025,
author = { Li, Haoshen and Wang, Yirui and Yu, Qinji and Zhu, Jie and Yan, Ke and Guo, Dazhou and Lu, Le and Dong, Bin and Zhang, Li and Ye, Xianghua and Wang, Qifeng and Jin, Dakai},
title = { { Metastatic Lymph Node Station Classification in Esophageal Cancer via Prior-guided Supervision and Station-Aware Mixture-of-Experts } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15972},
month = {September},
page = {359 -- 369}
}
Reviews
Review #1
- Please describe the contribution of the paper
The authors propose a Lymph Node (LN) prior-guided Attention Loss (ATTN-Loss) and a station-aware mixture-of-experts (SA-MoE) module to improve the accuracy of metastatic LN station classification. The experimental results demonstrate that the proposed methods bring performance improvements to MobileViTv2.
- 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 proposed methods and preprocessing workflow of first segmenting mediastinal LN-stations automatically are effective.
- The application for LN station classification of esophageal cancer is innovative to some extent.
- 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 baseline model, MobileViTv2, for the proposed method implementation may not the state-of-the-art for classification.
- Both source code and dataset are private. It is not easy to validate the results for reproducibility.
- 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
For revision, I would recommend: 1) Could the author provide the experimental results of MobileViTv3 (https://github.com/micronDLA/MobileViTv3, https://github.com/jaiwei98/mobile-vit-pytorch) to demonstrate whether the baseline model is the state-of-the-art? 2) “Authors are not allowed to change the default margins, font size, font type, and document style.” The default color for URLs in the LaTeX template and final publishing PDF file is blue (\renewcommand\UrlFont{\color{blue}\rmfamily}), not green or red. 3) The Reference section should follow the Springer Reference Style. If there are more than 6 authors, list the first author followed by “et al.”. The format for Springer proceedings is unique to others, please refer to the previous papers in the MICCAI proceedings. 4) The references should be updated to date; for example, the paper of MobileViTv2 [16] has been accepted by Transactions on Machine Learning Research (TMLR). 5) The authors should release the source code and/or dataset upon acceptance of the submission.
- 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 paper is innovative with its methodological approach and clinical application. However, the experimental validation is limited but convincing.
- 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
From a clinical perspective, the authors addressed the challenges of one-to-one LN matching between CT and pathology report; the main technical contribution lies in the shift of LN instance metastasis with LN-station class priors for LN classification
- 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 is well written. It presents a solid review of related work, and the explanation of the methods and results is sound. The ablation study on the effects of individual components, conducted using relevant backbones and methods, along with qualitative results, helps to clarify both the strengths and limitations of the proposed approach.
-
The paper includes a robust evaluation using the largest cohort to date for the study of EC (over 1,000 patients) with CT scans and pathology reports, evaluated via cross-validation.
-
The authors address the prevalent problem of LN metastasis by shifting the focus from LN instance-level analysis to pathology-confirmed station-level assessment, which holds high clinical relevance for preoperative planning. The proposed method proves effective, outperforming state-of-the-art approaches on this representative cohort.
-
Additionally, an LN-stations module is introduced to learn metastasis features across different LN stations. A novel LN attention loss is also proposed to enhance the framework’s robustness to suspicious LN regions.
-
- 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 lacks a thorough discussion of its limitations and potential directions for future work.
-
In the implementation details, the rationale behind fixing the voxel value to −1024 is unclear. Was this value chosen empirically, or is there a clinical justification for it, i.e., to capture locoregional LNs?
-
- 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.
- 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 link to the source code would be helpful
- 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?
It proposes a clinically relevant approach to reframe the LN metastasis problem by focusing on pathology-confirmed LN stations. Using the largest cohort to date, it also establishes new state-of-the-art baselines.
- 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 #3
- Please describe the contribution of the paper
The paper uses more than 1000 patients with esophageal cancer to assess LN station with pathology-confirmed labels. The study creates a model for LN station metastasis diagnosis. The method incorporates two key innovations: a LN prior-guided attention loss that explicitly guides the network to focus on suspicious LN regions, and a station-aware mixture-of-experts (SA-MoE) module that routes different LN-station groups to specialised experts.
- 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 problem is clinically relevant since it predicts variables in a pathology report. The dataset is very large although single centre. The proposed LN prior-guided attention loss is a novel contribution that effectively leverages automatically segmented LNs to focus the network’s attention on relevant regions. The station-aware mixture-of-experts module is a clever adaptation of the MoE concept. The ablation study is appreciated and supports the leading performance of the model.
- 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 LN segmentation method relies on a prerequisite model with 80% recall, but the impact of segmentation errors on the final classification performance is not discussed. The study appears to be single centre.
- 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.
- 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.
(6) Strong Accept — must be accepted due to excellence
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
This paper presents a novel approach to the clinically important problem of lymph node station metastasis classification in esophageal cancer. The method is technically sound, with innovations in both attention guidance and expert specialisation. The large dataset (1,153 patients) demonstrates clear performance improvements over state-of-the-art approaches. The ablation studies thoroughly validate the contribution of each component.
- 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
Author Feedback
We thank the reviewers for their highly constructive and positive feedback. The following are our responses to review comments.
Response to Reviewer # 1: Although the LN segmentation method may introduce some segmentation errors, it still provides valuable prior information about the locations of lymph nodes within each lymph node station. This helps guide the network to focus on the key regions and improves the classification performance of the lymph node stations. We acknowledge the importance of analyzing the impact of segmentation errors on the final classification results, and we will include this discussion in our future work.
Response to Reviewer # 2: We are aware that the current evaluation is limited to a single dataset, which may affect the generalizability of our findings. As part of our future work, we intend to conduct extensive experiments on multi-center datasets to better assess the performance and robustness of our approach. In addition, the selection of −1024 was based on clinical guidance rather than empirical tuning. It serves to mask out regions outside the lymph node stations, enabling the network to focus on the internal, clinically relevant areas.
Response to Reviewer # 3: We chose MobileViTv2 as the backbone to demonstrate the effectiveness of our proposed method. While there are newer state-of-the-art models for classification after MobileViTv2, we will evaluate the generality and performance of our approach across different backbone architectures as part of our future work. In addition, we sincerely appreciate the reviewer’s comments on LaTeX formatting and reference style. We will revise these aspects in the camera-ready version to ensure compliance with the formatting guidelines.
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