List of Papers Browse by Subject Areas Author List
Abstract
Acquisition of dynamic contrast-enhanced MR imaging with gadolinium-based contrast agents at multiple time points provides valuable diagnostic information. In breast MRI, dynamics of enhancement serve as key indicators for differentiating malignant from benign tumors. However, acquiring delayed-phase images requires extended scan times and could lead to patient discomfort and increased costs. Furthermore, some protocols acquire only early-phase images, limiting the ability to capture dynamics of enhancement over time. In this study, we propose an iterative deep neural network that sequentially generates post-contrast images using prior outputs. By synthesizing delayed-phase images at multiple time points from early acquisitions, the proposed network enables the temporal prediction of enhancement. We evaluate our approach using a breast MRI dataset consisting of images acquired at six time points, including the pre-contrast phase. The results indicate that the proposed method can approximate delayed-phase images from early-phase images, suggesting its potential to support abbreviated scan protocols in dynamic contrast-enhanced MRI. Our code is available at: https://github.com/goglxych97/iterU-Net.git
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/4195_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/goglxych97/iterU-Net.git
Link to the Dataset(s)
N/A
BibTex
@InProceedings{ChuWoo_Synthesizing_MICCAI2025,
author = { Chung, Woojin and Kang, Junghwa and Park, Ga Eun and Kim, Sung Hun and Nam, Yoonho},
title = { { Synthesizing Delayed-Phase Contrast-Enhanced Breast MR Images from Early-Phase Images Using an Iterative Deep Network } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15963},
month = {September},
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper presents an iterative deep neural network (iterU-Net) for synthesizing delayed-phase contrast-enhanced breast MR images from early-phase acquisitions. The key innovation lies in its ability to model temporal dependencies through sequential predictions: iterU-Net progressively generates time-series images by integrating sinusoidal time embeddings and a ConvLSTM block, which captures the physiological dynamics of contrast agent enhancement across multiple time points. Unlike conventional single-time-point synthesis methods, the iterative architecture refines outputs at each step using prior predictions, enabling accurate approximation of kinetic curves. To address training challenges, the authors introduced a warm-up stage to stabilize iterative prediction and foreground-aware random cropping to prioritize lesion regions, mitigating data imbalance.
- 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.
By combining iterative generation with time-series modeling, the model captures the physiological evolution of contrast agent enhancement through time embeddings and the ConvLSTM module, enabling multi-timepoint dynamic prediction. The warm-up training phase alleviates the instability of iterative prediction, while the lesion-aware random cropping improves the model’s focus on key lesion areas by balancing the distribution of lesion and background data.
- 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 iterative network requires a warm-up stage, which increases the complexity of training and computational cost, limiting its clinical application. The study uses a small dataset (198 individuals), which may not represent various clinical cases, and requires validation with a larger, more representative dataset. The evaluation is primarily based on signal intensity curves, lacking clinical assessment. The reliance on warm-up training limits the model’s scalability.
- 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.
(2) Reject — should be rejected, independent of rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
the iterative network requires a warm-up stage, increasing training complexity and computational costs, which limits its clinical applicability, especially in resource-constrained environments; the small dataset may not represent diverse clinical cases, requiring validation with a larger, more diverse dataset; the evaluation focuses on signal intensity curves rather than clinical assessments, limiting the model’s real-world clinical value; and the reliance on warm-up training restricts its scalability and adaptability in different settings. These factors hinder the model’s practical use and generalizability, leading to a lower score despite its technical feasibility.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Reject
- [Post rebuttal] Please justify your final decision from above.
Thank you for the authors’ response. While most of the issues were addressed, I remain unconvinced on the key points.
Review #2
- Please describe the contribution of the paper
The authors proposed an innovative method based on an iterative U-Net to address an interesting and relevant problem: synthezing late phase contrast enhanced breast MRI from early phase images. The approach iteratively predicts sequential time-series images, predicting the next image in the series iteratively and yields kinetic curves that closely approximate the real data. This application could lead to shortening scans, saving times and costs.
- 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 study is an innovative solution to an innovative application that could have real patient benefits.
They perform an ablation study on both the training scheme and architectural adaptations, and show good performance in comparison to an established baseline approach. The iterative nature makes it applicable despite variations in the acquisition that could be present in real-world clincial data.
- 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 evaluation would be stronger if it included a more directly clinically relevant task. Slightly more consideration of how these data can be used in practice would be appreciated. Expanding the approach to multi-vendor, multi-center data would be important for the future,
- 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?
It is an interesting problem, the solution seems to be appropriate and yields good results, although further evaluation would be appreciated.
- 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.
Although the authors rebuttal only acknowledges the concerns of the reviewers rather than trying to address any of them, this is probably a limitation of the short rebuttal period. Therefore, I decide to side with accepting this paper as I believe it is an interesting contribution to solving an interesting problem with potential clinical benefits. I hope that the reviewers concerns and suggestions can be addressed in a follow-up work.
Review #3
- Please describe the contribution of the paper
This paper proposes iterU-Net, an iterative deep learning model that synthesizes delayed-phase DCE-MRI images from early-phase acquisitions. The goal is to reduce scan time and patient burden by predicting future contrast-enhanced images, enabling better characterization of lesion enhancement patterns without requiring full dynamic scans.
- 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 proposed iterU-Net enables the synthesis of delayed-phase DCE-MRI images from early-phase inputs, which can reduce scan time and patient burden in clinical settings. 2.The iterative architecture effectively models temporal dynamics by using previous outputs and time embeddings, making it suitable for capturing enhancement kinetics over multiple time points. 3.Quantitative results show that the method significantly reduces enhancement error compared to baseline models, while maintaining similar or better PSNR and SSIM scores. 4. The model incorporates a diverse combination of loss functions—including L1 loss, SSIM, perceptual loss, and gradient difference loss—to preserve structural, perceptual, and contrast features in the synthesized images. 5. The use of foreground-aware random cropping ensures better lesion representation during training, helping to address class imbalance in the data.
- 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 training process requires a separate warm-up stage using ground-truth images, which adds complexity and makes the model harder to train end-to-end.
-
The iterative nature of the network increases computational cost and inference time, making it less efficient than direct single time-point prediction models.
-
The study is conducted on a relatively small dataset (198 cases, with only 38 used for testing), which may limit the generalizability of the results to broader clinical scenarios.
-
The evaluation primarily focuses on image similarity metrics and kinetic curve correlations, lacking clinical validation such as radiologist assessments or diagnostic accuracy studies.
-
- 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.
(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 presents a novel iterative deep learning framework (iterU-Net) for synthesizing delayed-phase breast DCE-MRI images from early-phase acquisitions. This approach is clinically meaningful as it addresses the important problem of reducing scan time while preserving the temporal dynamics critical for diagnostic decision-making. The proposed method demonstrates clear technical contributions, including time embedding, ConvLSTM integration, a well-structured warm-up training strategy, and a multi-component loss design. Quantitative results show improvements in enhancement error and kinetic curve fidelity compared to baseline methods.
However, the paper lacks clarity on external validation and generalizability, as the dataset is limited in size and scope. Additionally, the absence of code/data sharing or explicit reproducibility support limits immediate usability by the broader community. While the method is well-explained, clinical validation (e.g., radiologist evaluation) is missing, which would further strengthen the paper’s impact.
- 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.
Although point 2 (regarding the inefficiency of the iterative structure) was not directly addressed, the rebuttal is overall well-written and effectively responds to the main concerns.
Author Feedback
We sincerely thank all reviewers for their time and thoughtful feedback. We are grateful and encouraged by the reviewers’ recognition of the potential novelty and clinical relevance of our approach. We also appreciate the reviewers’ valuable suggestions and concerns, and we have done our best to address them thoroughly in the responses below:
(1) Reproducibility (R1, R2, R3) We fully acknowledge that the absence of explicit reproducibility support may hinder the immediate usability of our work. In response, we will make the full source code publicly available upon acceptance. To facilitate reproducibility, we will provide a script that enables the entire training process including the warm-up stage to run with a single command line.
(2) Further Evaluation: Dataset and Clinical Validation (R1, R2, R3) We agree that the limited size of our dataset, and agree that clinical validation would enhance the impact of our work. We will highlight the importance of future evaluation on larger, multi-vendor, multi-center cohorts in the revised Discussion. In line with R1’s suggestion regarding clinical relevance, we plan to extend our method to analyze DCE kinetic curve patterns, which may offer more interpretable outputs. Given the constraints of the rebuttal process, we are unable to implement this extension at present. However, we will include additional characterization of our dataset to help address this concern.
(3) Concerns about Warm-up Stage (R2, R3) We appreciate the R3’ concern regarding the added complexity introduced by the warm-up stage. We would like to clarify that it is a training-only strategy designed to improve stability and early convergence. It comprises approximately 30% of the training process time, requires no additional computational resources beyond the main training stage, and is excluded entirely from inference stage. Importantly, our main contribution (iterative prediction to capture temporal enhancement dynamics) is independent of the warm-up stage. Upon acceptance, we will release a public repository with an end-to-end implementation, including the warm-up stage, that can be executed with a single command line.
Once again, we sincerely thank the reviewers for their constructive feedback.
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.
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 #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