List of Papers Browse by Subject Areas Author List
Abstract
Brain tumor segmentation is a crucial task in medical imaging that involves the integrated modeling of four distinct imaging modalities to identify tumor regions accurately. Unfortunately, in real-life scenarios, the full availability of such four modalities is often violated due to scanning cost, time, and patient condition. Consequently, several deep learning models have been developed to address the challenge of brain tumor segmentation under conditions of missing imaging modalities. However, the majority of these models have been evaluated using the 2018 version of the BraTS dataset, which comprises only 285 volumes. In this study, we reproduce and extensively analyze the most relevant models using BraTS2023, which includes 1,251 volumes, thereby providing a more comprehensive and reliable comparison of their performance. Furthermore, we propose and evaluate the adoption of Mamba as an alternative fusion mechanism for brain tumor segmentation in the presence of missing modalities. Experimental results demonstrate that Transformer-based architectures achieve leading performance on BraTS2023, outperforming purely convolutional models that were instead superior in BraTS2018. Meanwhile, the proposed Mamba-based architecture exhibits promising performance in comparison to state-of-the-art models, competing and even outperforming Transformers. The source code is publicly released alongside the benchmark developed for the evaluation: https://github.com/AImageLab-zip/IM-Fuse.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/0747_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/AImageLab-zip/IM-Fuse
Link to the Dataset(s)
https://www.synapse.org/Synapse:syn51156910/wiki/627000
BibTex
@InProceedings{PipVit_IMFuse_MICCAI2025,
author = { Pipoli, Vittorio AND Saporita, Alessia AND Marchesini, Kevin AND Grana, Costantino AND Ficarra, Elisa AND Bolelli, Federico},
title = { { IM-Fuse: A Mamba-based Fusion Block for Brain Tumor Segmentation with Incomplete Modalities } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15967},
month = {September},
page = {225 -- 235}
}
Reviews
Review #1
- Please describe the contribution of the paper
The authors reproduce and evaluate key brain tumor segmentation models under missing modality scenarios using the BraTS2023 dataset (1,250 volumes). The study proposes the use of Mamba as an alternative modality fusion mechanism with missing modality.
- 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.
-
Comprehensive evaluation of existing methods on the large-scale BraTS2023 dataset under missing modality scenarios.
-
Introduction of a novel Mamba-based Interleaved Fusion Block for effective multimodal fusion in brain tumor segmentation.
-
Extensive experimental validation demonstrating the effectiveness of the proposed approach on BraTS2023, with detailed comparisons to state-of-the-art models.
-
Public release of source code, including the implementation of the proposed method and the reproduction of baseline models.
-
Showed from results, when dataset is larger, transformer-based architectures outperform convolution-based methods.
-
- 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 performance improvement is marginal compared to the baselines. From the table, I can see that methods like mmFormer has similar good performance and less parameters.
-
Some parts are difficult to understand, for example, in equation 2, what does FFN, LN, MSA mean?
-
- 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.
- 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?
This paper provides extensive benchmarking on BraTS2023 and further proposes a MEMBA-based architecture that achieves performance comparable to transformers. Although the improvement is marginal, it offers valuable insights to the field.
- Reviewer confidence
Somewhat confident (2)
- [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 presents a novel deep learning segmentation network for brain tumors in the presence of missing imaging modalities. The model is built upon a combination of Mamba and convolutional architectures. The authors evaluate their approach on the BraTS 2023 dataset, comparing its performance against several existing segmentation methods. Notably, the proposed network achieves competitive performance while maintaining a smaller parameter size and lower GFLOPs compared to other 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.
This paper proposes a new segmentation framework incorporating an Interleaved-MFB block. The authors provide comprehensive evaluations across multiple segmentation tasks, including scenarios with missing modalities. Additionally, the paper presents a well-organized and publicly available implementation on GitHub, which enhances the reproducibility and accessibility of the work.
- 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 “Preliminaries” section in the methods part seems less essential and could be streamlined. To better highlight the authors’ contributions, it would be more effective to present the Mamba block visually (e.g., with a diagram), rather than explaining it solely through text.
Since the compared methods were not originally designed for missing modality scenarios, the authors should consider fine-tuning those baselines for the missing modality setting—especially using the same loss functions introduced in the proposed method. This effort could be viewed as an additional contribution, particularly if the authors aim to benchmark fairly under consistent conditions. Alternatively, comparisons could be restricted to full-modality cases to avoid potential bias.
Additionally, this paper presents a segmentation approach distinct from both SegMamba and the rank-1 method from the original BraTS 2023 leaderboard. Citing and comparing with these relevant works would strengthen the paper’s positioning and completeness. For the authors’ reference, I have attached both papers. [1]Ferreira, André, Naida Solak, Jianning Li, Philipp Dammann, Jens Kleesiek, Victor Alves, and Jan Egger. “How we won brats 2023 adult glioma challenge? just faking it! enhanced synthetic data augmentation and model ensemble for brain tumour segmentation.” arXiv preprint arXiv:2402.17317 (2024). [2]Xing, Zhaohu, Tian Ye, Yijun Yang, Guang Liu, and Lei Zhu. “Segmamba: Long-range sequential modeling mamba for 3d medical image segmentation.” In International Conference on Medical Image Computing and Computer-Assisted Intervention, pp. 578-588. Cham: Springer Nature Switzerland, 2024.
- 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
The authors should consider including qualitative analysis of the results, such as example segmentation cases, to more intuitively demonstrate the superiority of their proposed method. Visual comparisons can effectively complement quantitative metrics.
To further strengthen the paper, the authors are encouraged to evaluate their method on additional datasets. Demonstrating generalizability across different benchmarks would significantly enhance the overall quality and impact of the work
- 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 well-written and presents methodological innovations that contribute to the field of medical image segmentation.
The provided code is well-organized and supports reproducibility.
The proposed method is compared against a range of existing approaches, demonstrating its relative performance.
- 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
Review #3
- Please describe the contribution of the paper
This paper addresses the important and practical challenge of brain tumor segmentation under missing modality conditions. While most previous works in this area rely on the relatively small BraTS2018 dataset, this study conducts a comprehensive reproduction and evaluation of existing methods using the much larger BraTS2023 dataset. In addition, the authors propose the adoption of a Mamba-based architecture as an alternative fusion mechanism to improve segmentation performance in the presence of incomplete modality 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.
- Clarity and Structure: The paper is well-organized and clearly written. The methodological descriptions are easy to follow, and the figures and tables are well-presented to support the narrative.
- Comprehensive Benchmarking: The authors conduct a detailed review, reproduction, and evaluation of existing models for handling missing modalities in brain tumor segmentation. This benchmark on BraTS2023 fills an important gap in the current literature and provides valuable insight into the generalizability of various approaches.
- Novel Fusion Strategy: The proposed Mamba-based fusion mechanism offers a novel perspective, and experimental results demonstrate its competitive—sometimes superior—performance when compared to state-of-the-art transformer-based models. This adds a fresh angle to the discussion on modality fusion techniques.
- 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.
- One notable limitation of the paper is the absence of qualitative results, such as visualizations of the segmentation outputs. Given the nature of the task—brain tumor segmentation under missing modality conditions—visual comparisons are essential for assessing the practical effectiveness of the proposed model. Visualizations can help highlight differences in boundary delineation, robustness to missing modalities, and failure cases that may not be evident through quantitative metrics alone. Including such results would significantly enhance the interpretability and credibility of the model’s performance.
- 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?
This paper presents a well-structured and thoroughly evaluated study on brain tumor segmentation under missing modality scenarios. The authors provide a clear and comprehensive comparison of existing state-of-the-art methods using the large-scale BraTS2023 dataset, significantly improving upon prior evaluations that were mostly limited to BraTS2018. This benchmarking effort alone is a valuable contribution to the community, offering a more robust understanding of model performance in realistic settings.
Moreover, the proposed use of the Mamba-based fusion mechanism introduces a novel and promising direction for handling incomplete multimodal data. The experiments are extensive and carefully designed, demonstrating that the proposed approach performs competitively—often surpassing existing transformer-based models.
While the lack of qualitative visualizations is a minor shortcoming, it does not substantially diminish the impact of the work. The methodological clarity, sound experimental validation, and meaningful insights justify acceptance. The paper is likely to serve as a solid reference for future research on robust medical image segmentation under incomplete data conditions.
- 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
We sincerely thank the reviewers for the time dedicated to reviewing our work and for the valuable feedback they provided. We take this opportunity to clarify some of the weaknesses highlighted by the reviewers.
In the following, citations identified with * are local to the present document, others refer to the original submission.
R1 - Competitor not designed for missing modality scenarios We would like to clarify that, although with different strategies, all the selected competitors are specifically designed to handle missing modality scenarios and are all trained from scratch using the same splits (publicly released on GitHub). For each model, we employed the same preprocessing, augmentation, optimizer, scheduler, and hyperparameters as described in their respective original papers, with the exception of the number of iterations, which were scaled to ensure an equivalent number of epochs as in the original studies due to the increased number of training samples. Additionally, for each model, we evaluated on the test set the version that achieved the highest metric on the validation set, thereby minimizing the risk of selecting an overfitted model.
Instead, neither suggested competitor [1] nor [2] is designed to handle missing modalities in MRI brain tumor segmentation, making those methodologies out of scope with respect to our experimental setting. We will make this even more explicit in the camera-ready version of our manuscript.
R1, R2 - Qualitative analysis We already generated visualization images for qualitative analysis, we’ll do our best to include them in the camera-ready version of the paper while complying with page limits. We thank the reviewers for the suggestion.
R1 - Additional Datasets We are already working on an extension of this work, where we are including different multi-imaging modality datasets (e.g., [*3, *4]). We thank the reviewer for the suggestion.
Additionally, we would like to clarify that, prior to the original submission, we also evaluated our method on the BraTS2018 dataset. However, we chose to report only the results on BraTS2023 in the manuscript for the following reasons:
(i) BraTS2023 is an extension of BraTS2018 and thus provides a more comprehensive and up-to-date benchmark. (ii) Most transformer-based competitors underperform on BraTS2018 due to its limited data size. (iii) Space limitations prevented us from including an additional table duplicating the structure of Table 1 for BraTS2018 results. (iv) Performance results of many competing methods on BraTS2018 are already publicly available in prior work [21, 23, 29].
R2 - Section Difficult to Understand [R2] In the camera-ready version of the manuscript, we explicitly defined terms like FFN, LN, and MSA used within equations. We thank the reviewer for the valuable feedback.
References [*1] Ferreira, André, Naida Solak, Jianning Li, Philipp Dammann, Jens Kleesiek, Victor Alves, and Jan Egger. “How we won brats 2023 adult glioma challenge? just faking it! enhanced synthetic data augmentation and model ensemble for brain tumour segmentation.” arXiv preprint arXiv:2402.17317 (2024). (https://doi.org/10.48550/arXiv.2402.17317)
[*2] Xing, Zhaohu, Tian Ye, Yijun Yang, Guang Liu, and Lei Zhu. “Segmamba: Long-range sequential modeling mamba for 3d medical image segmentation.” In International Conference on Medical Image Computing and Computer-Assisted Intervention, pp. 578-588. Cham: Springer Nature Switzerland, 2024. (https://doi.org/10.1007/978-3-031-72111-3_54)
[*3] Hernandez Petzsche, M. R., de la Rosa, E., Hanning, U., Wiest, R., Valenzuela, W., Reyes, M., … & Kirschke, J. S. (2022). ISLES 2022: A multi-center magnetic resonance imaging stroke lesion segmentation dataset. Scientific data, 9(1), 762. (https://doi.org/10.1038/s41597-022-01875-5)
[*4] Sergios Gatidis et al. Automated Lesion Segmentation in Whole-Body FDG-PET/CT, 2022. (https://doi.org/10.5281/zenodo.6362493).
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