List of Papers Browse by Subject Areas Author List
Abstract
Chemotherapy is the standard first-line treatment for lung cancer, and cellular death is an inevitable consequence of the process. However, current methods lack high-throughput, label-free approaches for accurately assessing cell death, and existing techniques struggle to capture cellular heterogeneity, complicating the prediction of lung cancer prognosis. Therefore, we propose frequency vision Mamba (FViM) for label-free cell death pathway prediction in lung cancer chemotherapy. Specifically, we introduce multi-dimensional optical time-stretch imaging flow cytometry (OTS-IFC) to capture high-throughput, multi-dimensional cell images under various cell death states. To effectively extract key features that are highly indicative of cellular heterogeneity, we propose FViM that integrates modeling remote dependencies of Mamba alongside frequency domain analysis of Fourier Transform. FViM first employs the frequency guided enhancement (FGE) module to enhance cellular detail features in the high-frequency domain, while reinforcing global contextual features in the low-frequency domain. The enhanced features are then processed through the Mamba-based visual state space block, which models the intricate relationships between different visual states, achieving a holistic prediction of cell death states. Experimental results demonstrate that FViM outperforms existing state-of-the-art (SOTA) methods. Notably, FViM successfully predicts cell death pathways in response to increasing cisplatin concentrations, demonstrating its potential for effective and promising applications in lung cancer chemotherapy. Our code is available at https://github.com/yzygit1230/FViM.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/3318_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/yzygit1230/FViM
Link to the Dataset(s)
N/A
BibTex
@InProceedings{YeZha_FViM_MICCAI2025,
author = { Ye, Zhaoyi and Wei, Shubin and Mei, Liye and Weng, Yueyun and Geng, Qing and Wang, Du and Lei, Cheng},
title = { { FViM: Frequency Vision Mamba for Label-Free Cell Death Pathway Prediction in Lung Cancer Chemotherapy } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15970},
month = {September},
page = {234 -- 244}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper aims to improve cell death pathway prediction in the context of lung cancer therapy by analyzing flow cytometry images with deep learning. Specifically, the authors appear to have developed a novel multi-dimensional optical time-stretch flow cytometry (OTS-IFC) system for capturing high-throughput, label-free intensity and phase-contrast images—though I cannot evaluate the technical specifics of the imaging system in detail, as it lies outside my area of expertise.
On the modeling side, the paper introduces FViM, a new architecture that combines Vision Mamba for efficient sequence modeling with FFT-based modules to jointly capture local texture and global semantic features. The dataset comprises approximately 8,000 cultured NSCLC cells, spanning three phenotypic conditions (control, autophagy, apoptosis) and cultures with varying concentrations of cisplatin, a chemotherapy drug.
The authors conduct a comparative study against state-of-the-art models, demonstrating strong performance, and include a careful ablation study to justify the architectural choices.
To evaluate the prediction of cell death pathways, UMAP is employed to project deep features into a 2D space, visually illustrating how cell populations—control, autophagy, and apoptosis—shift in response to increasing drug concentration. The gradual movement of clusters toward the apoptosis region is further supported by a corresponding increase in the proportion of apoptotic cells as the drug concentration rises. Finally, Grad-CAM is used to reveal how the network adapts its focus to increasing drug concentration.
- 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 is well-written, and the relevance of the novel developments is supported by the promising results.
- 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.
On cell-death pathways prediction : while Grad-CAM provides insights into the regions the model attends to, the specific mechanism through which it supports the authors’ hypothesis requires further clarification. It would be helpful to elaborate on how the visualizations from Grad-CAM directly validate the model’s predictions.
The statistical power of the analysis could be improved, as only a single train/validation/test split is used. It would be helpful to clarify which dataset was used for the results reported in Table 1. I assume these are test set results, given that the validation set is likely used for model selection and tuning, but this should be explicitly stated.
Reproducibility could also be improved, as some important implementation details are missing, particularly regarding the exact parameters and configurations used for the different models. Providing this information—either in the main text or supplementary materials—would greatly enhance the replicability of the work.
- 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?
The paper presents a compelling combination of optical innovation and deep learning tailored to relevant biomedical task. The contributions are novel and promising, but the paper would be stronger with more rigorous evaluation and reproducibility details.
- 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 proposes a novel method that predict the cell death states and cell death pathway of multi-dimensional optical time-stretch imaging flow cytometry (OTS-IFC) data. Besides, they introduce Fourier Transform to guide Mamba model to focus much on cellular detail features, which improve the classification performance.
- 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 topic is interesting. Assessing cell states and cell death pathway automatically is important for evaluating lung cancer chemotherapy and may promote other biological researches. This paper designs a Mamba-based method to solve this problem, which is a novel application.
- It sounds reasonable to introduce Fourier Transform to help Mamba to capture fine-grained details. The ablation studies also show the effectiveness of this module.
- The result in section 3.5 is interesting. It shows that the proposed method could distinguish different cell death pathway in an unsupervised manner.
- 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.
- In the introduction, this paper does not mention the previous method for analyzing OTS-IFC images. They only said “current implementations primarily focus on intensity-based cell image analysis…” without any citation.
- The results in Table 1 show that the performance improvement of FViM is minor. Since its improvement compared to MedMamba is less than 0.5% in all metrics. I don’t think this is a statistically significant improvement in a small dataset that only contains thoudsands of images. You may consider to use cross validation to increase the credibility of the results.
- Fig 2 (c) shows the Grad-CAM results that the model identifies key cellular patterns associated with different cells. However, it lacks discussion about the biological insight, such as the relation between the model’s attention and cell death pathways.
- 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
- How do you get the results in Fig 2 (b)? Is it the ground-truth or model prediction? If it is no ground-truth, how to evaluate the performance of cell death pathway prediction?
- 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 major factor that let me accept this paper is its interesting application. The weakness is the innovation and improvement of the proposed model is limited.
- 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 frequency-vision mamba for label-free cell death pathway prediction in lung cancer chemotherapy. Results on public datasets show state-of-the-art performance.
- 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.
They propose a novel architecture, with extensive ablation, analysis, and comparision.
- 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 model is only evaluated on private datasets, it will be good to see the performance on public benchmarks.
- 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.
(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?
Please see the comments related to weakness and strength.
- 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
Author Feedback
We sincerely appreciate the reviewers for their valuable recognition of our work and insightful suggestions, which have greatly enhanced the quality of our manuscript. Below we provide point-by-point responses to address all raised concerns.
Previous works [R3] We have added citation of previous intensity-based cell image analysis. e.g. [1, 2].
Dataset [R1, R2, R3] To the best of our knowledge, this work represents the first integration of a multi-dimensional OTS-IFC system with deep learning approaches for label-free cell death pathway prediction. We plan to further expand the dataset scale to contribute to the research community. Our study addresses two tasks: (1) cell death state recognition and (2) cell death pathway prediction. The experimental results presented in Table 1 were evaluated on the testing set of our cell death state recognition dataset.
Results [R2, R3] 1) Performance of cell death pathway prediction [R3] The quantitative results in Fig. 2(b) represent FViM’s predicted cell death distributions, demonstrating a dose-dependent response consistent with established cisplatin pharmacology - showing characteristic transition from autophagy priming at lower concentrations (7.5-15μM) to apoptosis dominance at higher doses (30-60μM), evidenced by the progressive decline in viable cells (73%→34%) accompanied by initial autophagic activation followed by apoptotic escalation [3, 4]. While this overall progression aligns with classical pharmacological responses, we observe subtle discrepancies in autophagic cell proportions that may reflect the model’s confusion between control and early autophagic states due to their morphological similarities. Future refinements will incorporate temporal tracking and additional biomarkers to better resolve these transitional phenotypes.
2) Grad-CAM [R2, R3] The Grad-CAM visualizations are in Fig. 2 (c) demonstrates that FViM effectively identifies distinctive morphological signatures associated with different cell death pathways: normal cells maintain regular contours, autophagic cells exhibit characteristic size expansion and cytoplasmic accumulation, while apoptotic cells present typical structural disintegration (as can be directly observed in Fig. 1 (a). These attention patterns not only correspond to well-established biological hallmarks of cell death but also provide compelling visual evidence that FViM’s decision-making process is grounded in biologically meaningful features. Importantly, this is quantitatively supported by Fig. 2 (b), FViM successfully captures the dose-dependent progression from autophagy to apoptosis at increasing cisplatin concentration, further validating its capability towards cell death pathway prediction.
Reproducibility [R2, R3] To ensure fair comparison, all comparative methods maintained their original architecture while adopting identical training hyperparameters (e.g., learning rate, batch size, and epochs) as detailed in the Implementation Details section. The link to the open-source code for FViM will be included in the camera-ready version to ensure reproducibility of this study.
[1] Weng, Y., et al.: Typing of acute leukemia by intelligent optical time-stretch imaging flow cytometry on a chip. Lab on a Chip 23(6), 1703–1712 (2023) [2] Zhou, J., et al.: Imaging flow cytometry with a real-time throughput beyond 1,000,000 events per second. Light: Science & Applications 14(1), 76 (2025) [3] Chen, L., et al.: Mitochondria-targeted cyclometalated iridium (III) complexes: Dual induction of a549 cells apoptosis and autophagy. Journal of Inorganic Biochemistry 249, 112397 (2023) [4] Lu, Y., et al.: ph-responsive, self-assembled ruthenium nanodrug: Dual impact on lysosomesand dna for synergistic chemotherapy and immunogenic cell death. Small 20(24), 2310636 (2024)
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