List of Papers Browse by Subject Areas Author List
Abstract
Pathologic complete response (pCR) prediction for breast cancer patients undergoing neoadjuvant chemotherapy (NAC) is crucial for optimizing treatment strategies. Nowadays, an increasing number of studies focus on predicting NAC response using preoperative imaging, and with the advancement of deep learning, different modalities of imaging and other clinical data can be effectively integrated to provide more comprehensive information. However, existing deep learning methods primarily focus on multimodal fusion or longitudinal modeling but often suffer from inadequate feature focus and overlook specific treatment effects. To address these limitations, we propose a novel multimodal-learning framework LMF(Longitudinal MRI-Clinical Multimodal Fusion) that enhances feature extraction and explicitly models treatment-induced imaging changes. Our method consists of two key components: (1) Molecular-Aware Deformable Attention (MADA), which integrates molecular subtype information with MRI features and refines spatial representations via deformable cross-attention mechanism; and (2) Treatment-Aware Longitudinal Modeling (TALM), which incorporates treatment embeddings to capture NAC-driven feature variations. The model is trained and evaluated on the ISPY-2 dataset, using pre- and post-NAC DCE-MRI alongside clinical data. Experimental results demonstrate that our approach outperforms existing methods, confirming that MADA effectively enhances feature extraction while TALM strengthens longitudinal modeling. These findings highlight the potential of integrating multimodal feature refinement with treatment-aware temporal modeling for improved pCR prediction. Our code is available at https://github.com/martin-bro/LMF.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/2501_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/martin-bro/LMF
Link to the Dataset(s)
N/A
BibTex
@InProceedings{MaDin_Longitudinal_MICCAI2025,
author = { Ma, Dingrui and Cao, Jiawei and Cheng, Hao and Zhou, Dan and Liu, Jianping and Zhang, Xiaofeng and Wu, Kaijie and Gu, Chaochen and Guan, Xinping},
title = { { Longitudinal MRI-Clinical Multimodal Fusion for pCR Prediction in Breast Cancer } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15974},
month = {September},
page = {326 -- 335}
}
Reviews
Review #1
- Please describe the contribution of the paper
The study focuses on prediction pathologic complete response from neoadjuvant chemotherapy using both pre- and post- treatment MRI clinical information such as hormonal receptor status.
- 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 strengths include:
- incorporating both receptor status and treatment information in a novel framework.
- solid ablation experiment.
- 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.
Weaknesses include:
- How is the approach clinically useful given that you use both pre and post treatment MRI. To have the post treatment information, means that is to late to change anything about the given treatment.
- the post treatment information is very important for the task, and when removed the sensitivity is only 0.48 meaning we will only be able to succesfully predict pathologic complete response in <50 % subjects. That will not be clinically useful.
- since there are so many public datasets in this space, your study needs to show independent validation.
- 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
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.
(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?
See balance of 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
Review #2
- Please describe the contribution of the paper
The paper proposes methods to exploit molecular and longitudinal imaging information to enhance the pCR prediction performance of deep learning models.
- 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.
- Exploiting molecular information and longitudinal images are meaningful and important for pCR prediction.
- The MADA and TALM modules are novel.
- The results are promising.
- 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 whole framework is too complex. Although the idea of introducing molecular information and longitudinal images is interesting, the way the authors achieve these targets is questionable.
- There are other important molecules involved in breast cancer, such as ER, PR, and Ki-67. The authors only use HER2, which could be limited.
- It seems that only pre-operative and one post-operative images are exploited at one time. Is it feasible to exploit images from multiple post-operative time points?
- The authors discussed in the introduction that “Early pCR prediction via preoperative imaging, such as DCE-MRI [17], offers the potential to optimize treatment plans and reduce overtreatment”. Exploiting images after treatment seems to be contradictory to this pre-operative prediction target.
- 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
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 idea is interesting. The results are promising. More clarifications on the methods are needed.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Accept
- [Post rebuttal] Please justify your final decision from above.
All my comments have been satisfactorily addressed. I hope that the authors will continue refining the method to enhance its clinical applicability further.
Review #3
- Please describe the contribution of the paper
This study enhances the accuracy of pCR prediction by integrating MRI and clinical information, which holds significant clinical value for guiding treatment strategies in breast cancer patients.
- 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 main strengths of this work include:
(1) Effective utilization and fusion of multi-temporal MR information.
(2) A well-designed integration and guidance strategy for incorporating clinical information into feature extraction, particularly through the inclusion of treatment plans.
(3) The proposed Dynamic Spatial Focus (DSF) module demonstrates an efficient design concept that potentially improves computational efficiency.
- 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 major limitations of this manuscript are as follows:
(1) The description in Section 2.1 (Overall Structure) is somewhat unclear and disorganized. A concise and clear overview is recommended, with detailed descriptions deferred to subsequent sections.
(2) The DSF module is only conceptually proposed, and its actual impact should be visualized and validated in the Results section. Its effectiveness remains uncertain without visual or quantitative evidence. Moreover, it is possible that the final dynamic spatial focus may not align with the tumor region.
(3) The set of SOTA (state-of-the-art) models used for comparison is relatively limited. The performance, especially AUC scores, is not particularly outstanding; in some prior works in the medical imaging domain, AUC values have exceeded 0.8.
(4) In Figure 1, the meaning of “DTP” is unclear — was this a typo?
(5) In Section 2.3, the acronym “TFF” needs to be defined. Also, the content in this section does not align well with the corresponding modules in Figure 1.
(6) The Methods section lacks a clear description of the final prediction module, resulting in an incomplete model pipeline.
- 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
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?
Writing quality and methodological innovation are the core aspects to be considered in the evaluation of this manuscript.
- 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
We sincerely thank all reviewers for their constructive comments and positive feedback on the novelty of our work. Below, we address the key concerns in detail.
The clinical utility of post-NAC MRI (Q1-1, 2; Q2-4): Most existing studies still focus on predicting pCR based on imaging prior to surgery. Since pCR is confirmed via postoperative pathology, imaging-based prediction remains clinically meaningful—it can help guide surgical decisions, such as whether to proceed with breast-conserving surgery or potentially avoid surgery altogether. Our study aligns with this practical objective, and in our context, “early” refers to making a decision before surgery, rather than waiting for postoperative histology. We would like to clarify that the ablation study cited did not remove post-NAC images; rather, it removed TALM, while retaining both pre- and post-treatment inputs. This result supports the effectiveness of incorporating treatment information. We recognize the limitations of relying solely on post-NAC imaging. A more impactful direction is to assess treatment response during NAC and adjust therapy accordingly. Our work explores this direction by integrating treatment embeddings into the longitudinal modeling. Moreover, we introduce a delta prediction branch in TALM to simulate post-NAC features from pre-NAC imaging and treatment data. This component will be a key focus of future work.
External validation and limited molecular features (Q1-3; Q2-2): We are currently conducting a multi-center study, which was still under preparation at the time of submission. Among public NAC+MRI datasets—DUKE, ISPY1, ISPY2, and NACT—we selected ISPY2 due to its relatively large cohort, inclusion of treatment, and axial-view DCE-MRI format, which matches our internal data. Additionally, using public data facilitates code and model release for community reproducibility. We acknowledge that ISPY2 only provides HR/HER2. In contrast, our internal dataset includes a more comprehensive panel of biomarkers such as ER, PR, HER2, and Ki-67, which we plan to incorporate into future models.
Timepoints selection (Q2-3): Due to cost and logistical constraints in clinical practice, patients typically undergo MRI at only 2–3 timepoints during NAC. We therefore focus on modeling pre- and post-treatment endpoints. However, our framework is general and can be extended to imaging modalities with higher temporal resolution, such as ultrasound. We are actively exploring this direction in follow-up studies.
Module analysis (Q2-1; Q3-2, 3): Although the effectiveness of each module has been demonstrated through ablation studies, we acknowledge that further quantitative and visual analyses are needed to better explain the internal mechanisms. This will be addressed in future work. On performance, we note that for complex tasks like pCR prediction using unannotated data, it is challenging to achieve very high AUC. We are actively refining our architecture to push performance beyond an AUC of 0.8, a goal aligned with clinical applicability. We also believe that our exploration of treatment-aware modeling provides a useful foundation for future research in personalized therapy response prediction.
Writing clarity and figure-text consistency (Q3-1, 4, 5, 6): We thank the reviewer for their comments on writing and clarity. We agree that Section 2.1 currently mixes architectural overview with implementation details, and we will restructure this section for better readability. We also appreciate the correction that the modules labeled “DTP” in Figure 1 should be “DFP”, and we will revise the figure accordingly. The “TFF” module is a cross-attention mechanism similar in design to MQF; both are shown in Figure 1B and distinguished by color. The final prediction head, which concatenates features and passes them through an MLP, is illustrated in the figure but was not elaborated in the text due to its standard nature. We will clarify this to ensure completeness.
Meta-Review
Meta-review #1
- Your recommendation
Invite for Rebuttal
- 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
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Reject
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
N/A
Meta-review #2
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
N/A
Meta-review #3
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
N/A