List of Papers Browse by Subject Areas Author List
Abstract
The pursuit of decision safety in clinical applications highlights the potential of transparent methods in medical imaging. While concept-based models offer local concept explanations (instance-level), they often neglect the global decision logic (dataset-level). Moreover, these models often suffer from concept leakage, where unintended information within soft concept representations undermines both interpretability and generalizability. To address these limitations, we propose Concept Rule Learner (CRL), a novel framework to learn Boolean logical rules from binary visual concepts. CRL employs logical layers to capture concept correlations and extract clinically meaningful rules, thereby providing both local and global interpretability. The results from two tasks demonstrate that CRL achieves competitive performance with existing interpretable methods while improving generalizability to out-of-distribution data. The code of our work is available at https://github. com/obiyoag/crl.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/1569_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/obiyoag/crl
Link to the Dataset(s)
N/A
BibTex
@InProceedings{GaoYib_Learning_MICCAI2025,
author = { Gao, Yibo and Zhou, Hangqi and Gao, Zheyao and Wang, Bomin and Gao, Shangqi and Wang, Sihan and Zhuang, Xiahai},
title = { { Learning Concept-Driven Logical Rules for Interpretable and Generalizable Medical Image Classification } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15960},
month = {September},
page = {293 -- 302}
}
Reviews
Review #1
- Please describe the contribution of the paper
1-The authors proposed a novel framework that learns Boolean logical rules from binary visual concepts to model concept correlations to address the issue of concept leakage.
2-The proposed method not only delivers concept explanations for individual predictions but also extracts decision rules for the entire datasets.
- 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.
1-The paper presents CRL, a novel framework that learns Boolean logical rules from binary visual concepts to model concept correlations while mitigating the issue of concept leakage. This approach offers a unique perspective on integrating local and global interpretability for medical image classification.
2-The paper is well-structured. The experimental section is comprehensive, evaluating CRL on two medical image classification tasks and comparing it with some existing methods.
3-The paper demonstrates that CRL not only offers concept explanations for individual predictions but also extracts decision rules for the entire dataset. Additionally, the experimental results suggest that CRL exhibits superior generalizability to unseen data compared to other concept-based models.
4-The authors have provided the code for their work, which is essential for reproducibility and further research.
- 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.
1-The Introduction section could be improved by including a diagram that visually represents the paper’s motivation. The current description involves several concepts and limitations, and a diagram would enhance readability and understanding.
2-While the paper qualitatively describes the interpretability of CRL, it lacks quantitative metrics to assess its effectiveness. Additionally, the generalizability claim is based on results from a single out-of-distribution (OOD) dataset, which may not be sufficient to draw broad conclusions.
3-The experimental results show that CRL achieves competitive performance compared to other methods, but it is not always the best-performing model. A more detailed analysis of the performance differences and potential limitations of CRL would be beneficial.
4-The paper lacks ablation studies that investigate the impact of individual components of CRL, such as the number of logical layers or the choice of binary concepts.
- 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 has provided an anonymized link to the source code, dataset, or any other dependencies.
- 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 proposed approach offers a unique perspective on integrating local and global interpretability for medical image classification, but the interpretability should be evaluated with more experiments.
- 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 authors present a paper on a deep learning framework that learns logical rules from visual inputs capable of extracting decision rules for entire datasets allowing for global interpretability. The approach is evaluated on two medical image classification tasks and compared to state-of-the-art models. The results show competitive performance and generalizability to unseen 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.
The major strength of the submitted manuscript lies in the formulation of the framework that combines logical layers to extract clinically meaningful rules with visual concepts and their in-depth evaluation on two clinical tasks. In particular, the proposed approach manages to leverage dataset-wide (“global”) contexts with instance (“local”) concepts for improved medical decision making. A key element is the usage of binary concepts an logical layers that are more robust towards concept leakage. Another strength is the rigorous evaluation on two different datasets in comparison to state-of-the-art methods showing superior or comparable results on the benchmark metrics.
- 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 does not exhibit major weaknesses.
- 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 has provided an anonymized link to the source code, dataset, or any other dependencies.
- 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?
My recommendation is based on the novelty of the contribution as well as the carefully designed experiments and comparisons to state-of-the-art methods to illustrate the validity of the proposed approach.
- 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 paper introduces a novel method, a Concept Rule Learner (CRL), which provides interpretable classification of medical images. First, concepts are predicted from raw images (ImageNet pretrained ResNet-34), then logical rules joining concepts are generated by boolean layers and finally the rules are associated with image class via a linear layer. The authors demonstrate comparable classification performance on the primary set and a much more favorable performance to SoTA methods when the learned model is transferred to a new dataset.
- 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 improves existing concept based methods (such as CBMs and CEMs) by introducing layers generating logical rules on top of the learned concepts. The rules are then used for prediction (as opposed to the bare concepts). The results demonstrate this approach has comparable for the primary dataset where it was trained and outperforms existing methods when transferred to a new dataset.
The paper is extremely well written with clear structure, language and clear mathematical formulations and schemes - I found it spotless…
The authors aggregate multiple emerging concepts and tricks to make learning of the model feasible, such as the use of the logical activation functions combined with a straight-through estimator.
The validation methodology is solid, uses three established datasets for two different medical image classification problems and compares the proposed method with a range of existing concept-based approaches in an apparently fair way.
The model structure provides a promising approach for interpretable and generalizable medical image classification.
- 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.
I did not find many weaknesses in the paper. However, it would be interesting to see, why the proposed model achieved high accuracy and not so good F1 score on both the concept and diagnostic metric.
Also, on the source dataset, the model performs better on the concept metric, than on the diagnostic metric. On the new dataset, the model outperforms competitors in almost all metrics - but the concept F1 score.
Insight into the precision/recall would be useful here, to see where the model lacks.
- 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 has provided an anonymized link to the source code, dataset, or any other dependencies.
- 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?
The paper presents a new promising method with clear methodology and a solid validation
- 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
N/A
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