List of Papers Browse by Subject Areas Author List
Abstract
Contrast-free AI myocardial infarction enhancement (MIE) synthesis technology has a significant impact on clinics due to its ability to eliminate contrast agents (CAs) administration in the current MI diagnosis. In this paper, we propose a novel cardiac physiology knowledge-driven diffusion model (CPKDM) that, for the first time, integrates cardiac physiology knowledge into cardiac MR data to guide the synthesis of high-quality MIE, thereby enhancing the generalization performance of MIE synthesis. The combining helps the model understand the principles behind the data mapping between non-enhanced image inputs and enhanced image outputs, informing the model on how and why to synthesize MIE. CPKDM leverages cardiac mechanics knowledge and MR imaging atlas knowledge to respectively guide the learning of kinematic features in CINE sequences and morphological features in T1 sequences. Moreover, CPKDM proposes a kinematics-morphology diffusion integration model to progressively fuse kinematic and morphological features for precise MIE synthesis. Evaluation on 195 patients including chronic MI and normal controls, CPKDM significantly improves performance (SSIM by at least 4%) when comparing with the five most recent state-of-the-art methods. These results demonstrate that our CPKDM exhibits superiority and offers a promising alternative for clinical diagnostics.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/0918_paper.pdf
SharedIt Link: pending
SpringerLink (DOI): pending
Supplementary Material: https://papers.miccai.org/miccai-2024/supp/0918_supp.pdf
Link to the Code Repository
N/A
Link to the Dataset(s)
N/A
BibTex
@InProceedings{Qi_Cardiac_MICCAI2024,
author = { Qi, Ronghui and Li, Xiaohu and Xu, Lei and Zhang, Jie and Zhang, Yanping and Xu, Chenchu},
title = { { Cardiac Physiology Knowledge-driven Diffusion Model for Contrast-free Synthesis Myocardial Infarction Enhancement } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
year = {2024},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15001},
month = {October},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
The authors propose a deep learning framework to generate virtual late gadolinium enhancement images. The framework includes separate modules to model cardiac motion and morphological features, these are subsequently combined to consider both features in generating images. They report image quality evaluation vs 5 previously published methods.
- 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.
Interesting method for a clearly clinically relevant problem with good comparison to literature with improved results.
- 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.
The writing style is not good and overstated the innovations and novelty of this work. Evaluation is not strong. The final model appears overly complex Limitations are not well discussed.
- 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.
- Do you have any additional comments regarding the paper’s reproducibility?
Will the code/models be made available?
- 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
There is more than one typo in the title MIE is not a common acronym It is not required to describe the work as novel and innovative so much. Please describe the methods and allow the reader to decide if it is novel There are too many acronyms Some steps are given unusual names which may confuse the reader. For example, “Cardiac mechanics-guided kinematics interpretation module” should be “Cardiac motion module” instead Sentences like: “VNE was published in Circulation (One of the best journals in cardiovascular disease research).” are not helpful. Just refer to the previous paper in a normal way. Implementation/model details are not clear. Is the system trained end-to-end? There are no statistical tests and evaluation is only focused on image quality metrics but not clinically relevant
- 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
Weak Reject — could be rejected, dependent on rebuttal (3)
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The difficulty in understanding the paper/methods and limited evaluation considered
- 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
Reject — should be rejected, independent of rebuttal (2)
- [Post rebuttal] Please justify your decision
I still find the innovation/novelty of the work are oversold, with very limited evaluation (no clinically meaningful metrics) to back it up. The complexity of the proposed three module set-up does not justify any incremental gains achieves
Review #2
- Please describe the contribution of the paper
The paper introduces a novel technique, the Cardiac Physiology Knowledge-Driven Diffusion Model (CPKDM), which is the first of its kind to integrate cardiac physiology knowledge into cardiac MR data. This integration guides the synthesis of high-quality Contrast-free AI Myocardial Infarction Enhancement (MIE) images, significantly enhancing the generalization performance of MIE synthesis. The CPKDM model leverages knowledge of cardiac mechanics and MR imaging atlas to respectively guide the learning of kinematic features in CINE sequences and morphological features in T1 sequences. Additionally, it proposes a kinematics-morphology diffusion integration model to progressively fuse kinematic and morphological features for precise MIE synthesis, demonstrating superior performance compared to recent state-of-the-art methods.
- 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.
Novel Formulation: Cardiac Physiology Knowledge-Driven Diffusion Model (CPKDM) The paper introduces a groundbreaking technique known as the Cardiac Physiology Knowledge-Driven Diffusion Model (CPKDM), which marks a significant advancement in synthesizing Contrast-free AI Myocardial Infarction Enhancement (MIE) images. This formulation revolutionizes the field by integrating cardiac physiology knowledge into cardiac MR data to guide the high-quality synthesis of MIE images. By leveraging knowledge of cardiac mechanics and MR imaging atlas to respectively guide the learning of kinematic and morphological features, CPKDM pioneers a new era in image synthesis and diagnostic accuracy, propelling the field towards safer and more effective clinical diagnostics.
Demonstration of Clinical Feasibility The paper demonstrates the clinical feasibility of CPKDM through robust experimentation and comparison with state-of-the-art methods. By synthesizing high-quality MIE images without the use of contrast agents, CPKDM exhibits superior performance across visual, imaging, and clinical metrics. This demonstration not only validates the effectiveness of CPKDM but also showcases its potential to offer a safer, faster, and more cost-effective alternative for clinical diagnostics. The comprehensive evaluation underscores the clinical feasibility of CPKDM and its promising implications for enhancing diagnostic accuracy.
- 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.
Lack of Adaptability to Highly Variable Individuals The CPKDM model may face limitations in adapting to highly variable individuals, potentially affecting its generalization capability. Existing methods, as highlighted in the paper, have struggled to demonstrate good performance when faced with highly variable individuals. Although CPKDM integrates cardiac physiology knowledge to enhance generalization, its performance in scenarios with extreme variability beyond the training data’s scope may pose a limitation. This could be further exacerbated in real-world clinical settings where patient variability is inevitable.
Inherent Complexity and Interpretability The integration of cardiac physiology knowledge with image data may introduce inherent complexity and reduce interpretability. While this approach enhances the generalization capability of MIE synthesis, the added complexity stemming from the fusion of cardiac mechanics knowledge and MR imaging atlas knowledge may limit the model’s transparency and interpretability, posing challenges in understanding the underlying principles guiding the synthesis process. This lack of interpretability can hinder its adoption in clinical practice and limit the ability to explain the model’s decisions and outcomes.
Robustness to Unforeseen Scenarios It is essential to consider the robustness of CPKDM to unforeseen scenarios, such as variations in imaging conditions or unanticipated physiological factors. As the model heavily relies on the integration of physiological knowledge with cardiac MR data, its robustness to unforeseen variations beyond the scope of training data may become a potential limitation, impacting its real-world applicability. A thorough understanding of how the model behaves in unforeseen scenarios and its resilience to unpredictable variables is crucial for its practical utility.
These weaknesses highlight the areas for further investigation and refinement in the formulation of CPKDM to ensure its adaptability to diverse clinical scenarios, enhance interpretability, and improve robustness in unforeseen situations. Additionally, addressing these limitations would contribute to the broader acceptance and implementation of CPKDM in the field of medical imaging and clinical diagnostics.
- 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.
- 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
1.The innovation brought forth by CPKDM introduces a commendable shift in cardiovascular imaging; however, addressing the inherent complexity and reducing interpretability should be prioritized. Enhancing transparency in the synthesis process and ensuring interpretability of the model’s decisions will foster greater trust and acceptance within the clinical community. Clear visualization of the model’s decision-making process and the integration of explainable AI techniques could greatly enhance the model’s interpretability, thus contributing to its real-world applicability and adoption. 2.In the cardiac kinematics explanation module, the myocardial segmentation in calculating myocardial motion is automatically segmented. The author should clearly introduce this. 3.In the evaluation metrics, the author should introduce the differences between image-level metrics and region-level metrics. In clinical metrics, is it only visual result evaluation? Lack of credibility.
- 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
Weak Accept — could be accepted, dependent on rebuttal (4)
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The key contributions, notably the innovative integration of cardiac physiology knowledge into cardiac MR data, highlight the paper’s pioneering efforts in addressing the need for contrast-free MIE imaging and its promising impact on clinical diagnostics. These contributions reinforce the significance and relevance of the paper, substantiating the recommendation for an elevated overall score.
- 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
Weak Accept — could be accepted, dependent on rebuttal (4)
- [Post rebuttal] Please justify your decision
The author’s rebuttal has addressed my questions, and I agree to accept it.
Review #3
- Please describe the contribution of the paper
The authors proposed a sophisticated machine learning model for predicting contrast-enhanced late gadolinium enhancement (LGE) imaging (the current clinical standard) using contrast-free cine and T1-mapping.
The topic and development is of high clinical relevance, as if it is proven clinically robust, this can significantly improve the workflow for scar assessment of myocardial infarction (heart attacks).
The methods are well presented, results are promising.
- 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.
High clinical relevance and impact.
Novel machine learning method, including three modules to improve the performance of image generators.
Clear presentation.
- 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.
The paper may have been strenghened by:
- details related to the data sets;
- clinical metrics: scar volume fraction;
- clinical validation: quantitative assessment of scar and transmularity.
- 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?
The paper reproducibility can be further strengthened by more details and specifications of data materials. For example,
- does 14625 refer to the number of sequences (e.g. short axis acquisitions) or image frames?
- Why the number of cine is significantly larger than of T1 maps, and significantly larger than of LGE?
- Has it been quality controlled? what is the QC procedure?
- 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
The clinical metrics results can be further strengthened by quantitative analysis.
The seminal paper on AI-VNE should be cited: Circulation. 2021;144:589–599
The VNE reproducibility paper should be cited: https://doi.org/10.1016/j.jocmr.2024.100956
- 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
Strong Accept — must be accepted due to excellence (6)
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Important topic on the AI-enhancement of contrast-free MRI using generative models. The topic is at the centre of discussion in the society for magnetic resonance research.
High clinical relevance and impact.
- 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
Author Feedback
We sincerely thank all the reviewers for your valuable comments. Special thanks to R1 for the strong accept and highlighting the “High clinical relevance and impact,” “Novel machine learning method,” and “Clear presentation.” We appreciate R3 for the weak accept and the praise of “groundbreaking technique” and “significant advancement”. Additionally, we thank R4 for your detailed review, noting the “Interesting method” and “improved results,” as well as for pointing out the issues in our writing and presentation. • Concerning issues of writing and presentation, such as 1)“uncommon and too many acronyms”[R4], 2)“typos in the title”[R4], 3)“unusual step names”[R4] and 4)“unhelpful sentences”[R4]. A: We have double-checked the entire manuscript and had it carefully proofread by a native speaker. 1) We have revised and reduced acronyms, for example, replacing “MIE” with “Myocardial Infarction Enhancement”. 2) We have corrected typos in the title. 3) We have corrected unusual step names, for example, replacing “Cardiac mechanics-guided kinematics interpretation module” with “Cardiac motion module”. 4) We have removed unnecessary sentences, for example, removing “VNE was published in Circulation”. • Concerning issues of implementation and model, such as 1)“Implementation/model details are not clear”[R4], including “number of data samples”[R1] and “Is the system trained end-to-end?”[R4]; 2)“Will the code/models be made available?”[R4]; and 3)“relevant literature”[R1]. A: Due to space constraints, some implementation/model details are included in the Supplemental Material of our submission. For example, each data sample consists of 25 CINE frames, 8 T1 images, and 1 LGE image. Additionally, other missing details, such as ‘the system is trained end-to-end’ will be included in the final manuscript. 2) We have open-sourced our code on GitHub. 3) We will include these VNE-related citations in the final manuscript. • Concerning issues of model design and performance, such as 1)“adaptability to highly variable individuals”[R3], 2)“robustness to unforeseen scenarios”[R3], and 3)“inherent complexity and interpretability”[R3, R4]. A: 1) Highly variable individuals are a recognized challenge in the field, and our knowledge-and-data-driven method is motivated to address this. Our method relies on myocardial mechanics and cardiac imaging atlases to flexibly infer myocardial function and cardiac structure, rather than solely on the training data. The experimental results also demonstrate the superiority of our method. 2) As mentioned above, our method also exhibits robustness to unexpected variations. This robustness is achieved by understanding and adapting to variations from both kinematic and morphological perspectives. 3）Our knowledge-and-data-driven method does not significantly increase complexity and enhances interpretability. On one hand, the introduction of knowledge only adds the myocardial strain calculation formula, rather than adding new layers of complexity. On the other hand, the model’s decision-making becomes more transparent. For example, the calculation of myocardial strain can directly show myocardial function. • Concerning issues of experimental analysis, such as 1)“quantitative analysis of clinical metrics”[R1, R3, R4], and 2)“differences between image-level and region-level metrics”[R3]. A: 1) The clinical quantitative results (i.e., Fig. 5 in the manuscript) are a correlation coefficient (R) of 0.91 and an intraclass correlation coefficient (ICC) of 0.95 for scar size (P<0.001), as well as an R of 0.85 and an ICC of 0.90 for transmurality (P<0.001), demonstrating a high consistency between the synthesized myocardial scars and LGE images. More quantitative results are available on our GitHub. 2) As stated in our experimental setup, image-level metrics assess the entire image quality, while region-level metrics focus specifically on the MI areas. All other minor comments will be carefully considered and addressed.
Meta-Review
Meta-review #1
- 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’
This paper is well-presented with significant technique contribution, and the concerns from the reviewers have been generally explained very well, even though I feel it would be more convicing to evaluate the model across centers to prove its generalization ability.
- What is the rank of this paper among all your rebuttal papers? Use a number between 1/n (best paper in your stack) and n/n (worst paper in your stack of n papers). If this paper is among the bottom 30% of your stack, feel free to use NR (not ranked).
This paper is well-presented with significant technique contribution, and the concerns from the reviewers have been generally explained very well, even though I feel it would be more convicing to evaluate the model across centers to prove its generalization ability.
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’
Most reviewers (2 of 3) agree to accept this paper.
- What is the rank of this paper among all your rebuttal papers? Use a number between 1/n (best paper in your stack) and n/n (worst paper in your stack of n papers). If this paper is among the bottom 30% of your stack, feel free to use NR (not ranked).
Most reviewers (2 of 3) agree to accept this paper.