List of Papers Browse by Subject Areas Author List
Abstract
Early prediction of Pathological Complete Response (pCR) in breast cancer patients undergoing Neoadjuvant Chemotherapy (NAC) is critical but challenging as early-stage morphological changes in tumors are subtle. Existing methods relying on long-term longitudinal images fail to extract lesion development trends in limited early-stage data. To address this challenge, we propose CALM (Cross-View Spatial-Temporal Aligning for Longitudinal Multimodal data), a framework designed to accurately predict pCR at an early treatment stage by effectively fusing subtle lesion features from multimodal longitudinal images, thereby compensating for the absence of later-stage data. To enable effective multimodal information integration and mitigate inter-modal discrepancies in acquisition views and spatial resolutions, we introduce a Physical-distance-weighted Cross-view Inter-modal Attention(PCIA) mechanism within the multimodal transformer block. Furthermore, to exploit complementary physiological information across modalities, we develop a Physiology-Gated Guidance Module(PGM), which leverages the high tumor-signal saliency of the DWI modality to guide cross-modal attention fusion among multimodal images. In addition, we propose a Trajectory Margin Contrastive Loss(TMCL) to explicitly model longitudinal pathological evolution trajectories on the latent feature manifold, encouraging the capture of both the directionality and the potential progression trend of therapeutic response. Experimental results on a collected longitudinal multimodal MRI dataset of 146 patients demonstrate that CALM achieves superior performance using only early-stage data, outperforming state-of-the-art (SOTA) methods, showing the potential of our heterogeneous multimodal fusion approach for early medical prediction tasks.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/2264_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/Anchor-Li/CALM
Link to the Dataset(s)
N/A
BibTex
@InProceedings{LiXin_PhysiologyGuided_MICCAI2026,
author = { Li, Xin AND Zhuang, Mingrui AND Cui, Mingke AND Liao, Zhixuan AND Zhang, Qiang AND Wang, Hongkai},
title = { { Physiology-Guided Cross-View Spatial-Temporal Aligning for Early Prediction of pCR in Breast Cancer from Longitudinal Multimodal MRI } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16895},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
The paper proposes CALM, a framework for early prediction of pathological complete response (pCR) to neoadjuvant chemotherapy in breast cancer using longitudinal multiparametric MRI at only two early timepoints (T0 pre-NAC and T1 after two cycles). Key innovations include: (i) a Physical-distance-weighted Cross-view Inter-modal Attention (PCIA) module that aligns axial DCE/DWI and sagittal T2WI features in physical space without resampling; (ii) a Physiology-gated Guidance Module (PGM) that uses DWI-derived attention to suppress background parenchymal enhancement (BPE) and focus on tumor regions; and (iii) a Trajectory Margin Contrastive Loss (TMCL) that encourages discriminative latent evolution between T0 and T1.
- 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.
PCIA elegantly bypasses anisotropic resampling artifacts by aligning multimodal features directly via physical distance-weighted sparse attention. PGM translates the physiological high tumor saliency of DWI into a practical gating mechanism for DCE scans. TMCL captures subtle morphological treatment shifts by formulating early longitudinal dynamics as a contrastive trajectory. The architecture directly addresses a critical clinical window where early pCR prediction can adjust neoadjuvant chemotherapy regimens.
- 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 framework implicitly assumes perfect spatial alignment across modalities and timepoints without quantifying residual registration errors. The evaluation relies entirely on a small single-center cohort of 146 patients without external validation or molecular subtype stratification. Crucial statistical rigor is missing, including confidence intervals, paired significance tests, and sensitivity analyses for hyperparameter thresholds.
- 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 has provided an anonymized link to the source code, dataset, or any other dependencies.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- 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/2026/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 — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The proposed mechanisms for geometry-aware multimodal fusion and longitudinal trajectory modeling are clinically intuitive and technically sound. The empirical validation remains fundamentally inadequate due to the small single-center cohort and absence of external testing. Missing robust baselines, statistical significance tests, and hyperparameter sensitivity analyses prevent a reliable assessment of the algorithmic claims. The unverified reliance on perfect cross-modal image registration critically undermines the real-world deployability of the geometry-dependent PCIA module.
- 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 studies early prediction of pCR to neoadjuvant chemotherapy in breast cancer using longitudinal multimodal MRI at two early time points. The main contribution is CALM, which combines: (1) a physical-distance-weighted cross-view inter-modal attention module to fuse axial DCE-MRI and sagittal T2WI, (2) a DWI-guided soft attention mechanism to emphasize lesion-relevant regions and reduce BPE interference, and (3) a trajectory margin contrastive loss to model early treatment response dynamics between T0 and T1.The method is evaluated on a 146-patient multimodal breast MRI dataset and shows improved performance over several baseline methods.
- 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.
Clinically relevant task. Early prediction of pCR from only early-stage MRI is important and well aligned with clinical decision making. Well-motivated multimodal design. The method addresses three realistic challenges together: multimodal heterogeneity, cross-view misalignment, and subtle early temporal changes. Interesting geometry-aware fusion. The PCIA module is a meaningful contribution, as it incorporates physical distance into cross-view attention instead of relying on naive resampling. Reasonable use of modality-specific prior knowledge. Using DWI to guide attention in DCE-MRI is intuitive and clinically motivated, especially for reducing BPE-related distraction. Helpful ablation study. The ablation results support that PCIA, PGM, and TMCL all contribute, with PCIA appearing particularly important.
- 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 dataset size relative to model complexity. The proposed framework is fairly complex, while the study includes only 146 patients, which raises concerns about overfitting and result stability. Insufficient implementation detail. The paper does not clearly describe the registration/alignment pipeline, temporal alignment, and several preprocessing details, which affects reproducibility and also the validity of the geometry-aware fusion claim. The “physiology-guided” claim is not fully validated. The DWI-guided attention is plausible, but the paper does not convincingly show that the learned soft attention truly corresponds to tumor regions rather than general discriminative regions. Limited clinical analysis. There is no subgroup analysis by subtype, treatment regimen, or BPE level, and no external validation, which weakens the translational claim. Ablation could be more targeted. The current ablation is useful, but it does not fully isolate whether PCIA is better than standard cross-attention or whether TMCL is better than a standard temporal contrastive loss.
- 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 has provided an anonymized link to the source code, dataset, or any other dependencies.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- 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/2026/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 — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The paper addresses an important clinical problem and proposes a coherent combination of cross-view fusion, DWI-guided attention, and temporal trajectory modeling. The overall idea is interesting and the ablation study is supportive.
However, my score is limited by the current level of validation. In particular, the dataset is relatively small for the complexity of the model, key implementation details such as alignment and preprocessing are under-specified, and the lesion-focused interpretation of the attention mechanism is not sufficiently verified. In addition, the evaluation lacks subgroup analysis and broader validation. Overall, I find the work promising, but the current empirical support is not yet strong enough for a clear acceptance recommendation.
- 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 proposes CALM (Cross-View Spatial-Temporal Aligning for Longitudinal Multimodal data), a framework designed to predict PCR at an early treatment stage (after only 2 cycles of NAC).
Its primary contributions include:
PCIA Mechanism: A cross-view attention module that aligns axial (DCE-MRI/DWI) and sagittal (T2WI) features using physical distance weights, preserving high-resolution details without the artifacts of volumetric resampling. PGM Module: A physiology-guided module that uses DWI saliency to create soft attention gates, focusing the model on hyper-cellular tumor regions and suppressing Background Parenchymal Enhancement (BPE) noise in DCE-MRI. TMCL Loss: A novel contrastive loss that explicitly models the direction and magnitude of the therapeutic response trajectory in the latent feature manifold
- 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.
Clinical Grounding: The PGM module’s ability to suppress BPE noise using physiological priors (DWI cellular density) demonstrates a deep understanding of breast MRI challenges.
Early Detection Focus: By explicitly modeling the evolution trajectory (TMCL), the model overcomes the difficulty of subtle morphological changes at the T1 phase, where tumors often haven’t shrunk yet.
Strong Empirical Results: The model achieves superior AUC (0.804) and Accuracy (0.774) compared to SOTA methods like MAP and LOMIA-T, maintaining a better balance between sensitivity and specificity.
- 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.
Computational Complexity: The 3D physical-distance-weighted attention and multimodal fusion involving multiple encoders increase the complexity of the pipeline. A more detailed discussion on inference time or memory overhead compared to simpler baselines would be beneficial
- 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.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- 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/2026/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 — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The paper proposes CALM, a framework for early PCR prediction that uses PCIA for cross-view fusion and TMCL for trajectory modeling. These innovations are well-tailored to the physics and physiology of MRI.
I have assigned a Weak Accept with a Confidence level of 1 because I am not a deep specialist in this specific sub-field of longitudinal multimodal fusion. To ensure a fair evaluation, I have taken a cautious approach with this initial score. I am fully prepared to upgrade my rating to a firm Accept depending on the authors’ response during the rebuttal phase and the consensus among more specialized reviewers.
- Reviewer confidence
Not confident (1)
- [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 the reviewers for their valuable and constructive feedback. Currently, we have adapted our method to another independently collected small dataset, yielding favorable initial outcomes. We intend to carry out more comprehensive validations in our future research. Additionally, we will provide a more detailed explanation of the data preprocessing steps, which were not thoroughly covered previously due to space constraints.
Meta-Review
Meta-review #1
- Your recommendation
Provisional Accept
- Please justify your decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. In case of an invitation for rebuttal, clarify which points are important to address in the rebuttal.
This paper proposes the CALM framework, integrating the PCIA module, PGM module, and TMCL loss to realize early prediction of breast cancer pCR using longitudinal multimodal MRI at two early timepoints, with evaluation conducted on a 146-patient single-center cohort. All reviewers recognize its clinically relevant task, innovative technical design (especially geometry-aware cross-view fusion), and supportive ablation results,
