List of Papers Browse by Subject Areas Author List
Abstract
Automated CT report generation plays a crucial role in improving diagnostic accuracy and clinical workflow efficiency. However, existing methods lack interpretability and impede patient-clinician understanding, while their static nature restricts radiologists from dynamically adjusting assessments during image review. Inspired by interactive segmentation techniques, we propose a novel interactive framework for 3D lesion morphology reporting that seamlessly generates segmentation masks with comprehensive attribute descriptions, enabling clinicians to generate detailed lesion profiles for enhanced diagnostic assessment. To our best knowledge, we are the first to integrate the interactive segmentation and structured reports in 3D CT medical images. Experimental results across 15 lesion types demonstrate the effectiveness of our approach in providing a more comprehensive and reliable reporting system for lesion segmentation and capturing. The source code is publicly available at https://github.com/yanniangu/ISRG-CT-MICCAI2025.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/1326_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/yanniangu/ISRG-CT-MICCAI2025
Link to the Dataset(s)
N/A
BibTex
@InProceedings{GuYan_Interactive_MICCAI2025,
author = { Gu, Yannian and Lei, Wenhui and Chen, Hanyu and Zhang, Shaoting and Zhang, Xiaofan},
title = { { Interactive Segmentation and Report Generation for CT Images } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15964},
month = {September},
page = {272 -- 282}
}
Reviews
Review #1
- Please describe the contribution of the paper
The core contribution of the paper is a multitasking network that takes both a CT image and a point (e.g. clicked by the user) as an input and outputs both the segmentation of a lesion and a granular classification of lesion attributes as defined in a structured reporting template.
- 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 core strength of the paper is the novelty of combining intereactive lesion segmentation with lesion attribute classification to improve the radiological workflow using structured reporting.
The effectiveness of this approach is shown in a thorough experimental setup that includes both public and private datasets with a diverse set of lesions including a zero-shot setting with unseen types of lesions.
Notably, the method outperforms single task approaches in both classification and segmentation of lesions highlighting the synergistic nature of the proposed architecture.
- 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 authors claim that the approach “significantly improves clinical workflow efficiency”. However, I see no experimental evidence to backup this claim e.g. involving radiologists or evaluating the reporting time and correctness of the proposed appraoch.
The zero-shot results ranging from 0% to 100% accuracy seem to suggest that the attributes are highly imbalanced and lack support. This raises the question if the test sample size is sufficient and if accuracy is the right metric for evaluation.
Since it is well established that the configuration of U-Net has a high impact on its performance the baselines for segmenation mus includ nnU-net.
What CNN is used as a baseline? Details are missing and this might help putting the classification performance in better perspective. At least a ResNet or DenseNet should be evaluated and more details provided.
While the granular assessment of lesions is an important reporting step, using these attributes for classifying the type of lesions might be a more impactful problem that is unfortunately not explore in this work.
- 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 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?
If the listed flaws in the experimental setup are adressed, I would recommend to accept the paper given the novelty and potential usefulness in interactive report of the proposed method.
- 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
This paper propose to combine interactive segmentation and report generation for CT images. Specifically, it utilizes the complementary nature of segmentation and report generation task by interacting the text and image tasks. Additionally, the paper uses the local cluster center instead of direct user click to refine the use prompt to better align with the feature space.
- 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.
- As user click can introduce randomness, refining the prompt, as proposed in the paper, can be very useful.
- Using segmentation and report generation as auxiliary tasks of each to help visual understanding.
- 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 using k-mean as prompt refinement helps reducing randomness, it could limit the amount of control the user have. It seems that the size of the local region can play a big role here but no ablation study is provided (please don’t perform the ablation during rebuttal, you can explain if you have the result already).
- In the segmentation performance evaluation, it is unclear how the input point prompt is obtained.
- It seems that the segmentation output at different click iteration can all be valid (good enough) as the use may not be able to make the segmentation to perform better with more clicks. If this is the case, how does the user know when to stop click and trust the generated report?
- 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 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 overall idea is interesting and the evaluation demonstrate the performance. However, as an interactive tool, it must be easy to use. In this sense, it is also important to demonstrate that the proposed point refinement does not “overwrite” the use prompt. Additionally, there must be a way to tell the user to stop if the segmentation mask at the later click iteration is not always better than the earlier iteration. Therefore, this paper’s acceptance should be conditioned on further explanations on the usability.
- 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 authors propose a novel interactive framework for 3D lesion morphology report generation. To the best of my knowledge, this is the first work to integrate interactive segmentation and structured report generation for 3D CT medical images.
The framework enables radiologists to generate detailed clinical reports with minimal point-based interactions during image review, significantly improving usability and efficiency.
Moreover, the authors introduce feature-space clustering–based point refinement and inter-task feature synergy strategies, which further strengthen the technical contribution of the work.
- 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 focus on interpretability is a valuable and timely perspective for the field.
Inspired by SAM, the authors propose an interactive approach to report generation, where user clicks are refined using a feature-space clustering–based point refinement method to further enhance the performance.
Additionally, an inter-task feature synergy mechanism is introduced, which facilitates bidirectional information flow between segmentation and report generation, allowing both tasks to mutually enhance each other’s performance.
- 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 mathematical formulation could be described more clearly to improve readability. For example, it is not explicitly explained how N(p) is derived, which makes it harder for readers to follow the overall methodology.
- 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 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.
(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?
Refer to strengths and weaknesses.
- 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
@Reviewer 1 Thank you for your comprehensive review and insightful feedback. We greatly appreciate your critical assessment that has identified important limitations in our current work. Regarding our claim that the approach “significantly improves clinical workflow efficiency,” we acknowledge this statement lacks proper clinical validation with radiologists. We will revise this assertion to more accurately reflect the theoretical nature of this potential advantage. Due to MICCAI submission rules, we cannot incorporate new experiments during the rebuttal period. Nevertheless, we are committed to addressing these limitations comprehensively in our future research, including conducting proper clinical validation studies, implementing more appropriate evaluation metrics for imbalanced attributes, strengthening our baseline comparisons with architectures such as nnU-Net, ResNet, and DenseNet, and exploring the promising direction of using attribute assessment for lesion classification as you thoughtfully suggested.
@Reviewer 2 Thank you for your thoughtful questions about our methodology. Our k-means prompt refinement operates only within small local regions around user points, preserving intent while reducing noise—a design choice balancing refinement with user control. For evaluation, we obtain input point prompts by simulating expert feedback, randomly selecting points from misclassified regions (both false negatives and positives) when comparing predicted masks with ground truth. Regarding when users should stop clicking, our current method lacks explicit stopping criteria—a limitation we acknowledge. From a design perspective, incorporating visual feedback on segmentation confidence and stability between iterations would help users determine when additional input yields diminishing returns. We appreciate these insights highlighting both the strengths and limitations of our current methodological approach.
@Reviewer 3 Thank you for pointing out this lack of clarity in our mathematical formulation. We acknowledge that our paper does not explicitly explain how the local feature window N(p) is derived, which makes it difficult for readers to follow our methodology. In our approach, N(p) represents a local neighborhood window centered at point p, defined as all pixels within a fixed radius r from p using Euclidean distance. We will revise the manuscript to clearly define this important component with its proper mathematical formulation and provide context for how this local window is utilized in our feature aggregation process. We appreciate this feedback which will help improve the readability and reproducibility of our work.
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