Abstract

Whole-body diffusion-weighted imaging (DWI) is a sensitive tool for assessing the spread of metastatic bone malignancies. It offers voxel-wise calculation of apparent diffusion coefficient (ADC) which correlates with tissue cellularity, providing a potential imaging biomarker for tumour re-sponse assessment. However, DWI is an inherently noisy technique requiring many signal aver-ages over multiple b-values, leading to times of up to 30 minutes for a whole-body exam. We present a novel neural network implicitly designed to provide high-quality images from heavily sub-sampled diffusion data (only 1 signal average) which allow whole-body acquisitions of ~5 minutes. We demonstrate that our network can achieve equivalent quality to the clinical b-value and ADC images in a radiological multi-reader study of 100 patients for whole-body and abdo-men-pelvis data. We also achieved good agreement to the quantitative values of clinical images within multi-lesion segmentations in 16 patients compared to a previous approach.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/3447_paper.pdf

SharedIt Link: pending

SpringerLink (DOI): pending

Supplementary Material: https://papers.miccai.org/miccai-2024/supp/3447_supp.pdf

Link to the Code Repository

N/A

Link to the Dataset(s)

N/A

BibTex

@InProceedings{Zor_EnhancedquickDWI_MICCAI2024,
        author = { Zormpas-Petridis, Konstantinos and Candito, Antonio and Messiou, Christina and Koh, Dow-Mu and Blackledge, Matthew D.},
        title = { { Enhanced-quickDWI: Achieving equivalent clinical quality by denoising heavily sub-sampled diffusion-weighted imaging data } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
        year = {2024},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15003},
        month = {October},
        page = {pending}
}


Reviews

Review #1

  • Please describe the contribution of the paper

    The paper demonstrates the use of deep learning to speed up whole-body DWI MRI sequencing by enhancing sub-sampled data of multiple b values. In this way, whole body imaging and the assessment of bone metastases can be assessed more widely at lower cost. The paper details a robust method that performs well compared to modern methods like DNIF.

  • 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 paper presents a novel loss function using predicted ADC map. This is likely very important to guide the training, given the large number of parameters in the model. In addition the performance is assessed over large number of anatomical regions, which adds to the impact of the paper and the robustness of the results for clinical 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.

    While the large scan range contains a large number of slices, which result in what appears to result in a large dataset, this comes from a small number of patients. Also the patients appear to come from a single center, so there is an open question as to how the results generalise.

  • Please rate the clarity and organization of this paper

    Very 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
    • Has any hyperparameter tuning been performed? I miss this detail.
    • Fig 1: In the lower block, it is unclear to me what happens between the down sample and up sample block. Can you clarify this? Is the green from the right of the downsample the same as the green from the left of the up sample?
    • Extra metrics such as SSIM would be useful in addition to RMSE.
    • In Fig 2 (right) The small very dense plots are very difficult to discern less than obvious trends. Please move this to its own figure so it can be bigger.
    • Limitations of the study are not discussed as far as I can see. This is important given the relatively small number of patients and the lack of an external validation center.
  • 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 paper details a new method to enhance fast whole body MRI sequences and this performs well in relation to the state of the art. The method is assessed in many anatomical regions. There is a question about generalisability given the small number of patients and the lack of an external validation.

  • 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 #2

  • Please describe the contribution of the paper

    The paper proposes a deep learning-based denoising method for whole-body DWI. The authors propose several key modifications compared to a baseline approach: 1) using multiple b-values as network input, 2) introducing spatial context by inputing neighboring slices, and 3) a loss function employing ADC maps derived from the denoised predictions. The authors provide extensive quantitative and qualitative results.

  • 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 methodology introduces novel elements.
    • The evaluation is comprehensive, covering various aspects of performance.
    • The method consistently outperforms the chosen baseline in the experiments.
  • 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 absence of an ablation study makes it unclear which specific modifications contribute to the performance improvement compared to the baseline.
    • The study only includes one comparison method, limiting the breadth of the evaluation and potential insights gained from alternative approaches.
  • Please rate the clarity and organization of this paper

    Very 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?

    I appreciate that the authors included a comprehensive overview of the data acquisition and network training parameters in the supplementary material.

  • 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

    In Chapter 1.2, the authors mention the “quantitative nature of DWI”. However, this statement could be misleading, as DWI itself is diffusion-weighted and does not inherently provide quantitative measurements. The quantitative aspect typically refers to the derived ADC maps from DWI data.

  • 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?

    I think it is novel and interesting to e.g. use the ADC maps to compute the loss. However, in my opinion the experiments and evaluations lack some clarity in attributing specific performance improvements to each adjustment made compared to the baseline method.

  • 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

    This paper describes a technique to reconstruct diffusion image data with a denoising CNN-based reconstruction, trained in a supervised manner by retrospectively undersampling the acquisitions. The method introduces several innovations over a previously published method (DNIF) and compares the two, along with a fully-sampled standard acquisition. The analyses include quantitative evaluations in segmented regions and qualitative reader assessment of quality

  • 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.

    Significance: the topic of accelerating body diffusion imaging is a growing area with potential clinical impact

    Innovative methods: the paper introduces three innovations over DNIF: simultaneous training of multiple b-values, a loss function term based on ADC calculation, and adding a 2D+ approach by including using adjacent slices in the model input. The 2D+ approach is likely responsible for the improved appearance in the slice direction evident in Figure 3.

    Organ-specific analysis: the use of segmentation to focus on specific anatomical analysis ROIs increases the relevance and is seen as a strength.

    The reader study design is good (with exception noted below) and adds to the study impact

  • 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 qualitative evaluation is done using both validation and test data, which can lead to over-estimation of the performance.

    It is not unclear if the multiple b-value training or the ADC-based loss innovation has an impact on performance relative to DNIF, or if this is just due to the inclusion of adjacent slices.

  • Please rate the clarity and organization of this paper

    Very 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.

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

    The data and methods are not shared, and thus reproducing this work would be challenging

  • 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) Table 1: please add the number of stages and total acquisition times, as these data are relevant to the claims about scan times. Also, the row describing “directions” is unclear and needs some explanation. 2) Does using a single direction instead of geometric averaging over multiple directions lead to artifacts, for example due to flow? 3) The standardization procedure (eq 1) needs further explanation. Is the decrease of signal with increasing b-values preserved? Please define the parameters (xi, mu, etc). 4) The term “Celescan” is fig 2 is not defined 5) Please specify the method for generating ADC from the multi-b data (i.e., iterative non-linear fitting or using a log transformation + linear fit)

  • 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?

    This is a well-written paper on a significant topic. The improvement over the prior method is modest but impactful.

  • 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 would like to thank the reviewers for their positive and constructing comments on our manuscript.

We agree with the reviewers and acknowledge that our dataset currently is mono-centric. To the best of our knowledge the way this dataset is acquired is unique. We are conducting efforts to increase our data diversity by including 200 more patients from our centre, and note that the protocol used by our scanners is consistent with that others have used for this vendor. Additionally, we aim to enrich our quantitative validation methodology with more well-established metric such as the SSIM and deep feature representation. In this paper, we focused on showcasing the precision of the quantitative values inside the areas of disease which is crucial for treatment monitoring.

Ablation studies to reveal more deeply how each change over the DNIF architecture improved our results is something we are also very interested in. However, due to the marginal differences in the quantitative image measures, their effect can only be meaningfully assessed in the qualitative reader study we performed and it is part of our future plans. From our current experiments we having noticed that the inclusion of spatial context through neighbouring slices in the training instead of a single slice improved the perception of contrast during image transition. Incorporating ADC calculation in the loss function resulted in more accurate values in the resulting ADC maps indicating it acts as a regularizer.

Furthermore, we apologise for some “packed” presentation of the results and the figures, it was done due to the page limit. We would like to provide further information, such as that our deep learning network underwent hyperameter optimization, the ADC map was calculated using a linear fit and we did not notice any artifacts from geometric averaging. The downsample and upsample blocks are independent and are only connected through a skip connection.




Meta-Review

Meta-review not available, early accepted paper.



back to top