Abstract

Accurate prediction of Cardiovascular disease (CVD) risk in medical imaging is central to effective patient health management. Previous studies have demonstrated that imaging features in computed tomography (CT) can help predict CVD risk. However, CT entails notable radiation exposure, which may result in adverse health effects for patients. In contrast, chest X-ray emits significantly lower levels of radiation, offering a safer option. This rationale motivates our investigation into the feasibility of using chest X-ray for predicting CVD risk. Convolutional Neural Networks (CNNs) and Transformers are two established network architectures for computer-aided diagnosis. However, they struggle to model very high resolution chest X-ray due to the lack of large context modeling power or quadratic time complexity. Inspired by state space sequence models (SSMs), a new class of network architectures with competitive sequence modeling power as Transfomers and linear time complexity, we propose Bidirectional Image Mamba (BI-Mamba) to complement the unidirectional SSMs with opposite directional information. BI-Mamba utilizes parallel forward and backwark blocks to encode longe-range dependencies of multi-view chest X-rays. We conduct extensive experiments on images from 10,395 subjects in National Lung Screening Trail (NLST). Results show that BI-Mamba outperforms ResNet-50 and ViT-S with comparable parameter size, and saves significant amount of GPU memory during training. Besides, BI-Mamba achieves promising performance compared with previous state of the art in CT, unraveling the potential of chest X-ray for CVD risk prediction.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/1149_paper.pdf

SharedIt Link: pending

SpringerLink (DOI): pending

Supplementary Material: https://papers.miccai.org/miccai-2024/supp/1149_supp.pdf

Link to the Code Repository

https://github.com/RPIDIAL/BI-Mamba

Link to the Dataset(s)

N/A

BibTex

@InProceedings{Yan_Cardiovascular_MICCAI2024,
        author = { Yang, Zefan and Zhang, Jiajin and Wang, Ge and Kalra, Mannudeep K. and Yan, Pingkun},
        title = { { Cardiovascular Disease Detection from Multi-View Chest X-rays with BI-Mamba } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
        year = {2024},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15005},
        month = {October},
        page = {pending}
}


Reviews

Review #1

  • Please describe the contribution of the paper

    The paper makes a significant contribution to the field of medical imaging by proposing a novel approach, the BI-Mamba model, for predicting cardiovascular disease risk from multi-view chest X-rays. The model outperforms established methods like ResNet-50 and ViT-S and offers superior performance while maintaining linear time complexity and saving significant GPU memory during training. These results highlight the potential of using chest X-ray imaging for accurate and efficient cardiovascular disease risk prediction.

  • Please list the main strengths of the paper; you should write about 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 introduction of the BI-Mamba model represents a novel approach to predicting cardiovascular disease risk from chest X-ray images. This model leverages state space sequence models and bidirectional information encoding, offers a unique solution to capturing long-range dependencies in multi-view chest X-rays. BI-Mamba demonstrates linear time complexity, making it more computationally efficient compared to traditional methods like ViTs, which have quadratic time complexity. This efficiency is crucial for handling high-definition inputs and optimizing memory usage during training. The BI-Mamba model outperforms established architectures like ResNet-50 and ViT-S in terms of predictive capability while achieving comparable parameter size. This superior performance shows the effectiveness of the proposed model in accurately predicting cardiovascular disease risk from chest X-ray images. By unlocking the potential of using chest X-ray imaging for cardiovascular disease risk prediction, the BI-Mamba model offers a low-dose, cost-effective, and accurate alternative to traditional CT scans. This could have significant implications for clinical practice by enabling early intervention and personalized health management strategies.

  • Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.

    While the paper compares the BI-Mamba model with ResNet-50 and ViT-S, it lacks a comprehensive comparison with a wider range of state-of-the-art models in the field of cardiovascular disease risk prediction from medical imaging data. The paper could benefit from a more in-depth analysis of the interpretability of the BI-Mamba model’s predictions. The paper does not extensively discuss the scalability and deployment aspects of the BI-Mamba model in real-world clinical settings.

  • 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 mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure reproducibility.

  • Do you have any additional comments regarding the paper’s reproducibility?

    N/A

  • Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html

    Methodological Innovation and Contribution (MIC):

    To enhance the methodological contribution, consider providing more detailed insights into the design choices, model architecture, and the rationale behind specific components of the BI-Mamba model.

    Clinical Applicability and Impact (CAI):

    While the BI-Mamba model shows promising results in predicting cardiovascular disease risk, further emphasis on the clinical applicability and impact of the model is essential. Consider discussing the potential implications of the BI-Mamba model in real-world clinical settings, including its integration into existing healthcare workflows, patient outcomes, and cost-effectiveness compared to traditional diagnostic methods.

    Addressing the clinical validation of the model using diverse patient populations and real-world data would strengthen the clinical relevance and applicability of the BI-Mamba model.

    Translation of Methodology to Clinical Practice:

    To facilitate the translation of the BI-Mamba model into clinical practice, provide more insights into the scalability, interpretability, and deployment considerations of the model.

    Discuss how the model can be integrated into clinical decision-making processes, potential challenges in implementation, and strategies for ensuring the model’s reliability and interpretability in a clinical setting.

    Health Equity Considerations:

    Evaluate the potential impact of the BI-Mamba model on health equity by discussing how the model can address disparities in cardiovascular disease diagnosis and treatment across different demographic groups.

    By addressing these aspects in your paper, you can enhance the methodological rigor, clinical relevance, and potential impact of the BI-Mamba model in advancing cardiovascular disease risk prediction and improving patient outcomes.

  • 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

    Accept — should be accepted, independent of rebuttal (5)

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    Considering the innovative methodology, methodological advancements, efficiency in handling high-resolution inputs, and superior performance in CVD risk prediction, accepting this paper for publication in MICCAI would contribute significantly to the field of medical imaging and advance the state-of-the-art in cardiovascular disease risk assessment from chest X-ray images.

  • Reviewer confidence

    Very confident (4)

  • [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed

    N/A

  • [Post rebuttal] Please justify your decision

    N/A



Review #2

  • Please describe the contribution of the paper

    A novel neural network architecture variant is proposed (BI-Mamba) and applied to classify cardiovascular disease on chest X-ray images. The method is compared with three state of the art methods and shows better results and computational performance.

  • Please list the main strengths of the paper; you should write about 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.

    -Clearly structured, content well explained -Novel method presented -Application clinically relevant -Rich analysis: compared to two state of the art methods, in classification performance and GPU usage, with four different resolutions, and comparing sinlge view vs. multi view.

  • Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.

    -Method [1] is not explained in Introduction. It is not well comparable because only the heart region is analysed (see also “constructive comments”) -Unclear, what the innovation is compared to Vision mamba (reference [22]), which also contains bidirectional SSMs -SWIN also claims linear complexity with image size, which could weaken the claim in the introduction that transformers generally have quadratic complexity. How this relates could be explained in 2.3 -Computing time not listed in results (“Analysis of computational efficiency”), instead repeated complexity statements from 2.3. Interesting in the results sections are the real numbers.

  • Please rate the clarity and organization of this paper

    Excellent

  • 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.

  • Do you have any additional comments regarding the paper’s reproducibility?

    N/A

  • Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html

    By projecting in two direction, you lose information from whole CT. It would be nice to read in the discussion or even show experimentally: would AUROC improve with more projections? Could result from [1] be reached which uses 3D information (of the heart only)?
    Results and values should not appear in the introduction.

  • 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

    Accept — should be accepted, independent of rebuttal (5)

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    Clarity, novelty of method, applicability in e.g. screening setting, results

  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed

    N/A

  • [Post rebuttal] Please justify your decision

    N/A



Review #3

  • Please describe the contribution of the paper

    The paper presents BI-Mamba, a novel model developed for predicting cardiovascular disease risk through the analysis of chest X-rays. BI-Mamba leverages a combination of forward and reverse information processing, enhancing its interpretive capability of the imagery. This approach positions it as more efficient than earlier techniques such as ResNet-50 and ViT-S, particularly when it comes to processing high-resolution X-rays with reduced memory requirements.

  • Please list the main strengths of the paper; you should write about 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 study unveils BI-Mamba, an innovative model that employs a distinctive fusion of forward and backward information processing to evaluate chest X-rays for predicting the risk of cardiovascular diseases, marking a pioneering method in this domain.

    It underscores the clinical viability of favoring chest X-rays over CT scans for assessing cardiovascular risks, presenting a safer and more attainable alternative for patient care, pivotal for broad clinical implementation.

    BI-Mamba’s performance evaluation is notably robust, surpassing current models such as ResNet-50 and ViT-S regarding accuracy and memory efficiency, underscoring its capability and promise for practical healthcare use.

  • Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.

    Despite its innovation, the BI-Mamba model has not undergone a direct comparison with other advanced models tailored for cardiovascular disease (CVD) risk assessment from chest X-rays, which obscures its comparative effectiveness.

    The discourse on how the BI-Mamba model might be clinically applied and woven into current healthcare processes is insufficient, a gap that is essential to bridge for its practical implementation.

    Furthermore, the paper falls short of exploring the potential challenges and constraints in decoding the model’s forecasts, particularly in instances where the chest X-rays exhibit nuanced or complex signs of CVD.

    Although the paper introduces an early input patch concatenation technique for analyzing multi-view chest X-rays, it lacks a comprehensive evaluation against other fusion methods to establish its benefits.

  • Please rate the clarity and organization of this paper

    Excellent

  • 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.

  • Do you have any additional comments regarding the paper’s reproducibility?

    Please provide the link to your code.

  • Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html

    Conduct a comparative analysis by benchmarking BI-Mamba against a range of state-of-the-art models that are also designed for CVD risk prediction from chest X-rays. This will help to clearly delineate the model’s relative performance and innovative contributions to the field.

    Expand the discussion on the model’s integration into clinical workflows, considering factors such as interoperability with existing systems, ease of use by healthcare professionals, and potential changes in workflow that the model might necessitate. Articulate a clear pathway for how BI-Mamba can transition from a research prototype to a clinical tool.

    Delve deeper into the interpretive aspects of the model’s predictions, focusing on complex cases that present with subtle or atypical radiographic features of CVD. Offer guidance on how clinicians should handle and understand the predictions in such cases, possibly by providing additional decision support tools or incorporating explainability features within the model.

  • 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

    Accept — should be accepted, independent of rebuttal (5)

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    The paper is innovative and well written.

  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed

    N/A

  • [Post rebuttal] Please justify your decision

    N/A




Author Feedback

We are grateful to the reviewers for acknowledging our contributions and considering our work as “a significant contribution to the field of medical imaging”. The reviewers’ questions mainly regard the method comparison, chest X-ray generation, method clarification, paper’s reproducibility, and clinical applicability. This response letter clarifies these questions.

  1. Comparison with SOTA cardiovascular disease risk prediction method in Chest X-ray [R1, R4]

Using Chest X-ray for cardiovascular disease risk prediction is a very new research topic and thus there is few related work. The most relevant work [1] on this topic appeared after the submission deadline of MICCAI 2024, which uses convolutional neural networks to analyze chest X-rays to predict 10-year risk for major adverse cardiac events. We have been contacting the authors for their implementation and datasets to set up experiments to compare their method’s performance with the proposed BI-Mamba. Comparison with this cutting-edge method will be included in the camera-ready version.

[1] Weiss, Jakob, et al. “Deep learning to estimate cardiovascular risk from chest radiographs: a risk prediction study.” Annals of Internal Medicine 177.4 (2024): 409-417.

  1. Two projected chest X-ray views from CT [R3]

Our study generated frontal and lateral chest X-rays to perform cardiovascular disease risk prediction, bearing in mind that these two views are the most commonly ordered screening tests in clinical practice. We acknowledge that with the inclusion of more projected chest X-rays to draw a closer proximity to CT, the prediction AUROC will presumably increase. However, this setting loses its connection with clinical practice.

  1. Clarification of the comparison method in CT [R3]

The Tri2D-Net method in CT [2] used a detector to extract the heart region from a chest CT volume and a three-branch network to combine features extracted from 2D slices in three orthogonal views (axial, sagittal, and coronal) for cardiovascular risk prediction. In this study, instead of limiting the ROI to the heart region, we preserved a complete chest X-ray for cardiovascular risk detection. Such implementation follows our observations that imaging features outside the heart, such as blood vessel problems, are also critical factors to determine cardiovascular risk.

[2] Chao, Hanqing, et al. “Deep learning predicts cardiovascular disease risks from lung cancer screening low dose computed tomography.” Nature communications 12.1 (2021): 2963.

  1. Open-source code for paper’s reproducibility [R1, R3, R4]

The link to the open-source code for BI-Mamba will be included in the camera-ready version to ensure the reproducibility of this study.

  1. Clinical applicability [R1, R3, R4]

The traditional method for cardiovascular prevention computes the atherosclerotic cardiovascular disease (ASCVD) risk score based on the demographic attributes, smoking history, blood pressure, cholesterol level, etc. Our proposed method represents a cardiovascular risk estimator applicable to the cohort of patients who undergo chest X-ray screen and whose measurements required to calculate the ASCVD risk score are not available.




Meta-Review

Meta-review not available, early accepted paper.



back to top