List of Papers Browse by Subject Areas Author List
Abstract
Focal liver lesions (FLL) are common clinical findings during physical examination. Early diagnosis and intervention of liver malignancies are crucial to improving patient survival. Although the current 3D segmentation paradigm can accurately detect lesions, it faces limitations in distinguishing between malignant and benign liver lesions, primarily due to its inability to differentiate subtle variations between different lesions. Furthermore, existing methods predominantly rely on specialized imaging modalities such as multi-phase contrast-enhanced CT and magnetic resonance imaging, whereas non-contrast CT (NCCT) is more prevalent in routine abdominal imaging. To address these limitations, we propose PLUS, a plug-and-play framework that enhances FLL analysis on NCCT images for arbitrary 3D segmentation models. In extensive experiments involving 8,651 patients, PLUS demonstrated a significant improvement with existing methods, improving the lesion-level F1 score by 5.66%, the malignant patient-level F1 score by 6.26%, and the benign patient-level F1 score by 4.03%. Our results demonstrate the potential of PLUS to improve malignant FLL screening using widely available NCCT imaging substantially. Code is availabel at https://github.com/alibaba-damo-academy/plug-and-play-diagnosis.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/1669_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/orgs/alibaba-damo-academy/plug-and-play-diagnosis
Link to the Dataset(s)
N/A
BibTex
@InProceedings{HaoJia_PLUS_MICCAI2025,
author = { Hao, Jiacheng and Zhang, Xiaoming and Liu, Wei and Yin, Xiaoli and Gao, Yuan and Li, Chunli and Zhang, Ling and Lu, Le and Shi, Yu and Han, Xu and Yan, Ke},
title = { { PLUS: Plug-and-Play Enhanced Liver Lesion Diagnosis Model on Non-Contrast CT Scans } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15974},
month = {September},
page = {466 -- 476}
}
Reviews
Review #1
- Please describe the contribution of the paper
The paper proposes a plug and play framework that uses segmentations and prior lesion predeictions to refine the diagnosis of focal liver lesions on non-contrast CT images. The framework consists of a hierarchical dual attention mechanism and graph-based prior reasoning and uses a otimization strategy based on three different loss terms to ensure consistency across clinical workflow stages.
- 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.
Usability: Tthe proposed framework is compatible with any aribtrary 3D segmentation (and prediction 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.
Limited applicability: The overall objective appears weak, as it primarily focuses on refining classification outcomes based on existing segmentations and prior predictions.
Novelty and Methodological Clarity: the optimization strategy, including patient-level considerations, appears similar to that in related work [27], reducing the distinctiveness of the contribution. The integration of GPR is neither adequately motivated nor clearly explained or discussed.
Comparative Analysis questionable: The authors claim that multi-task frameworks perform suboptimal due to competing objective gradients, and want to improve the results from those frameworks using both the segmentation and prior prediction. However, in the experimets the compare themselfs two pure segmentation networks (nnUnet and Mask2Former). It is unclear how prior predictions are obtained, undermining the claimed advantage over multi-task approaches.
Discussion and Limitations are missing: The manuscript lacks a dedicated discussion of the results and does not address the limitations of the proposed approach. Including these sections would help contextualize the findings, clarify the scope of applicability, and guide future research directions.
- 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 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
Clinical Motivation and HDA Design: While the Hierarchical Dual Attention (HDA) mechanism is motivated by clinical observations—specifically that radiologists zoom in and out when examining lesions—the implementation does not fully align with this rationale. The multi-scale pooling is applied to liver features rather than lesion features, and the model lacks an explicit encoding of the positional relationship between lesion and liver features. This discrepancy weakens the clinical justification for the proposed architecture.
Missing Reference: The statement that “the graph-based structure automatically learns to enhance relevant prior knowledge while suppressing unreliable predictions, enabling more robust feature enhancement compared to conventional fusion methods” requires appropriate citation. Please provide references to prior work that supports this claim.
Low Recall Performance: The proposed method consistently demonstrates low recall, indicating a high number of false negatives. This is a critical issue, particularly in clinical applications, and should be thoroughly discussed in the manuscript.
Figure 3 – Incomplete and Arbitrary Representation: Figures 3b and 3c appear arbitrary in their depiction of subclasses. While the dataset is divided into benign and malignant lesions, each with multiple subclasses, the figures present only two subclasses—and not consistently the same ones. A comprehensive visualization including all nine subclasses would provide a more complete and meaningful analysis.
Ground Truth Reliability: The manuscript mentions that a subset of cases was annotated by radiologists, which was subsequently used to regenerate results. However, the size of this subset, the regeneration process, and the method used to ensure the correctness of ground truth labels remain unclear. Clarification is needed to assess the reliability and validity of the annotations used in training and evaluation.
Figure Captions and Abbreviations: Please provide more informative figure captions and ensure that all abbreviations, labels, and symbols are clearly defined. This is essential for reader comprehension and reproducibility.
Imaging Modality Justification: The authors motivate their approach by focusing on non-contrast CT (NCCT) images rather than contrast-enhanced CT (CECT) or MRI. While a direct comparison of classification performance across modalities is beyond the scope of this work, it would be valuable to acknowledge this as a potential direction for future research.
- 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.
(3) Weak Reject — could be rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
While the direction of the work is relevant and potentially impactful, the manuscript in its current form lacks methodological clarity, rigorous evaluation, and sufficient novelty to merit acceptance. A more thorough justification of design choices, stronger comparisons, clearer presentation, and a deeper discussion of results and limitations would be necessary for future consideration.
- 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 proposed a plug-and-play enhanced liver lesion diagnosis model, which compatible with arbitrary 3D segmentation models. For this purpose, a hierarchical dual attention, a graph-based prior reasoning modules are exploited to enhance FLL 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.
This method can maximizes feature discrimination from the pre-detected regions of interest (ROIs). They effectively integrate intra-, inter-scale semantic fusion with liver and lesions. The experiment shows that applying PLUS module is effective for various 3D segmentaiton models. It has high utility in that it enhances the output of other segmentation models.
- 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 pre-trained model used in the experiment seems to be outdated, from 2021 to 2022. 3D segmentation is a rapidly developing field, and it is necessary to verify whether it works well even with the latest model. Discussion and conclusion are not sufficiently provided.
- 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 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?
Overall, the research is well organized.
- 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
Liver lesion classification as benign and malignant is a difficult task for the radiologist due to small variations in the Hounsfield Units. Deep learning based methods are under development which exploits the image features and the deep features which helps the doctors as an assistive tool in lesion classification. This paper is an attempt in this situation in which the authors explained the Focal Liver Lesions detection from Non-Contrast CT images. The paper content is reasonably well written.
- 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.
Technically the paper is well written with required enough information in each section. The content is reasonably convincing the reviewer. Identifying the FLL clinically is challenging and the AI can act as assistive solution, and the work is promising as it is evaluated on a vast dataset.
- 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.
Observations:
- Gerneral remarks
- Abstract, line 2: For improving patient survival,
- Please use the references in order.
- Introduction is too lengthy, can be reduced
-
There is no much details about the data collected, such as data sources (only two hospitals has been mentioned), no details of patient inclusion and exclusion criteria, data preprocessing steps, how anonymisation was performed on CT images, image acquisition protocols (mainly the parameters). It is not clear whether 8651 are CT volumes or 8651 images. This needs to be clarified. If it is 8651 volumes, what was the number of axial CT images in each volume? how much effort was involved for annotating each volume?
-
Annotation protocol: Your 8651 volumes can become a potential candidate for the prospective researchers through FAIR principles and open science practices. I appreciate the effort involved in annotations. But there is no clarity on how was the annotation performed? what tool was used, who defined the annotation protocol?, how the annotations were cross checked by the experts, how many of the annotations were corrected in second round?, what was the plane used for annotation (natural slice or MPR)?
-
Did your model produce the same result when tested with images acquired with different imaging protocols at multiple centers? How do you ensure the model generalizability when evaluated on multi-center data? There is no mention of data availability.
-
Performance metrics for optimal model on all data partitions is not clear. Did you do any failure analysis of incorrectly classified ones?
-
There is no mention on how the results can be reproduced from the model, subjective evaluation of the model, inter-observer variations and its suitability to use in clinical workflow.
- What is your opinion on using CECT and multi-phasic images in training and testing phases?
- Gerneral remarks
- 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
For results reproducibility, there is no mention of data and the AI model availability in the paper. If code is available also, it is not clear about data partitioning, data preprocessing, evaluating the model performance etc. Please refer to https://pubs.rsna.org/doi/10.1148/ryai.2020200029 for AI research in Medical imaging.
- 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 paper is worth to accept as the methodology and the results are convincing. The authors must address the review comments before presenting
- 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
Dear reviewers and meta-reviewers,
We would like to extend our sincere appreciation for your evaluation of our manuscript. We note most concerns relate to more detailed descriptions and clarifications. We will try to clarify these in the camera-ready version.
We sincerely appreciate your invaluable time and efforts once again.
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