List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Abdullah-Al-Zubaer Imran, Sen Wang, Debashish Pal, Sandeep Dutta, Evan Zucker, Adam Wang
Abstract
The increasing frequency of computed tomography (CT) examinations has sparked development of dose reduction techniques to reduce the radiation dose to patients. Optimal dose while maintaining image quality can be achieved through accurate and realistic dose estimates. Unfortunately, existing dosimetric measures are either prohibitively slow or heavily reliant on absorbed dose within a cylindrical phantom, thereby ignoring the impact of patient anatomy and organ radiosensitivity on effective dose. We propose a novel deep learning-based patient-specific CT organ dose estimation method namely, multimodal contrastive learning with Scout images (Scout-MCL). Our proposed Scout-MCL gives accurate and realistic dose estimates in real-time and prospectively, by learning from multi-modal information leveraging image (lateral and frontal scouts) and profile (patient body size). Additionally, the incorporation of an accurately modeled tube current modulation (TCM) enables Scout-MCL to learn realistic dose variations. We evaluate our proposed method on a scout-CT paired scan dataset and show its effectiveness on predicting diverse TCM doses.
Link to paper
DOI: https://link.springer.com/chapter/10.1007/978-3-031-16431-6_60
SharedIt: https://rdcu.be/cVD7g
Link to the code repository
N/A
Link to the dataset(s)
N/A
Reviews
Review #1
- Please describe the contribution of the paper
This paper provides an interesting approach to estimate the tube current modulation (TCM) map from scout images and patient size. The approach includes a novel contrast learning technique to include information from different sources (images and size profiles).
The contributions are clearly stated:
- Contrastive learning technique to include information from images and patient profiles.
- Estimation of TCM maps.
- Real time CT organ dose estimation from scout images and TCM maps.
- 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 approach includes different sources of information.
Excellent agreement with predicted TCM and reference organ doses.
Low computation time required.
- 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.
Some details on the TCM map generation are not clear:
What does DCT stands for in this case?
Authors describe that the doses are modeled as the weighted sum of single-view doses with weights coming from the tube current profile.
How are the weights calculated from the tube?
How is the contribution of each organ calculated for each view?
Was the training and test set of scans split randomly?
The augmentation was performed only in scale of the same TCM maps. Differences in the scouts would be recommendable.
How are the inconsistencies between scouts and CT scans palliated?
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Note, that authors have filled out a reproducibility checklist upon submission. Please be aware that authors are not required to meet all criteria on the checklist - for instance, providing code and data is a plus, but not a requirement for acceptance
The code is not available (although not required). The datasets are not available (although not required).
Experiment parameters and configurations were detailed. Please clarify how the training and test were split.
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://conferences.miccai.org/2022/en/REVIEWER-GUIDELINES.html
I believe this is a relevant paper for the community and proposes a novel methodology that is worth publishing. Please, consider the weaknesses already pointed out to improve the clarity of this work.
- Rate the paper on a scale of 1-8, 8 being the strongest (8-5: accept; 4-1: reject). Spreading the score helps create a distribution for decision-making
6
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The weaknesses of this paper rely on the clarity of the presentation but not on the methodology itself. I think it could be easily improved with minor cosmetic changes.
- Number of papers in your stack
3
- What is the ranking of this paper in your review stack?
1
- Reviewer confidence
Confident but not absolutely certain
- [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
This paper proposes a method to predict patient specific organ dose in CT acquisition. The method takes as inputs both image-like information : 2 orthogonal 2D scout CT images, the scan z range in form of a binary 2D image, a tube current modulation map and 1D patient size profiles derived from the two scout images (Dw profiles). The objective is thus to be able to estimate dose from only scout images for tubes with TCM capability, alleviating for the need of a CT scan, which would be a major step towards dose personalization in CT imaging, which is of great clinical importance. The authors contributions are :
- multimodality : both image-space and profile-space are embedded into compressed latent representation and combined for dose prediction
- contrastive learning is used to correlate learned features of both encoders from orthogonal views of the same patient while disentangling them from views of other patients
- generate a TCM map, a part I did not fully catch
- 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 clinical value is high.
- The idea to leverage contrastive learning between two profile views of a same patient to improve dose prediction is very interesting and quite original in the field, especially given the demonstrated improved performances
- 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.
Many ideas are combined (TCM map generation, contrastive learning in the profile domain,”multimodality”), which makes the reader wonder what is the core message of the paper. It is not clear if the gain between pure image-based prediction (Scout-Net) and “multimodal-based prediction” (Scout-MCL) is actually due to the multimodality, as there is not an ablation study with respect to other image inputs of the ScE branch that are used by Scout MCL (i.e. the TCM map).
More importantly, I wonder if the DwE is really useful as contrastive learning could also be done directly in the scout image domainI Struggle to understand to what extent this information is multimodal, given that the Dw profiles are extracted from the scout images. Unless I am mistaken (which could be the case given my non proficiency in CT technicalities), this is more here a combination of a projected representations the scout views rather than indeed modalities that are combined.
the whole method describing the generation of synthetic TCM maps from scout images is unclear. It is said that “we generated synthetic 2D TCM maps by fitting profiles from DCT basis images to TCM profiles as is shown in Fig. 2”, which shows only a result.
The paper is very technically oriented towards CT and thus requires expertise in CT that I unfortunately do not possess completely.
- Please rate the clarity and organization of this paper
Satisfactory
- Please comment on the reproducibility of the paper. Note, that authors have filled out a reproducibility checklist upon submission. Please be aware that authors are not required to meet all criteria on the checklist - for instance, providing code and data is a plus, but not a requirement for acceptance
The lack of technical description regarding TCM map generation (the DCT fitting part of the paper that I missed) is a minus. Maybe it’s very common CT knowledge that I just do not possess.
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://conferences.miccai.org/2022/en/REVIEWER-GUIDELINES.html
- why not perform contrastive learning directly in the scout image domain ?
- I disagree with the unsourced comment that “Existing contrastive learning methods are heavily reliant on the availability of extremely large training sets. Especially in medical imaging, training data are limited making the contrastive learning methods less effective.”.
- section 4.3 should be renamed results
- Rate the paper on a scale of 1-8, 8 being the strongest (8-5: accept; 4-1: reject). Spreading the score helps create a distribution for decision-making
6
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
This is a very technical, CT oriented paper that proposes to leverage pseudo multimodality and contrastive learning in an original way that I think would find an audience at MICCAI.
- Number of papers in your stack
2
- What is the ranking of this paper in your review stack?
2
- Reviewer confidence
Somewhat Confident
- [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 proposes a deep learning approach for CT dose estimation from scout images and size profile. It is based on multi-modal self-supervised learning (with contrastive-learning (CL)), followed by dose learning from the deep self-supervised representations. A Tube Current Modulation (TCM) modeling is also proposed as an intermediate step to improve the dose estimation.
- 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 is well written and the method seems sound, with an appropriate use of CL for learning meaningful representations.
- 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.
An ablation study is performed only for the CL and some augmentation, not for other design choices such as the inclusion of body size and TCM in the pipeline. The dataset is rather small and it is not clear whether it originates from a single or multiple centers.
- Please rate the clarity and organization of this paper
Satisfactory
- Please comment on the reproducibility of the paper. Note, that authors have filled out a reproducibility checklist upon submission. Please be aware that authors are not required to meet all criteria on the checklist - for instance, providing code and data is a plus, but not a requirement for acceptance
The data does not seem public and no mention of sharing the code.
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://conferences.miccai.org/2022/en/REVIEWER-GUIDELINES.html
Abbreviations are used before their definitions, also notations (l \in {1,2,…L+1}) Some abbreviations should be reminded in Fig. 1. Some are defined after the figure. In Introduction, “in medical imaging, training data are limited …” It is not always true, there are a lot of unlabelled data for instance in histopathology. In introduction, sentence to fix: “… in order to maximize and minimize respectively…” In section 2., it is mentioned that the scout and Dw should learn similar representations. First it is not the image and Dw that lear representations, but even more I don’t think the representations should be similar. The image contains much more information than the Dw. More information could be provided for the model description, such as the number of filters, filter size, initialization for the 1D convolution. Equation (3) should be better motivated. It is also not well defined. Should it sum for all organs, or maybe one loss for each organ? Why are there L+1 organs? Also the notation l is used in the previous equation for the contrastive loss, and also for “leaky”, maybe other notations should be used. It is not clear whether the data originates from a single institution or more (plural in 4.1). Below table 1 (4.2), “l \in L” should be “l \in {1,…L}” or L+1 that’s not clear from before. In Table 1 it is strange to name the model …CL (noCL)
- Rate the paper on a scale of 1-8, 8 being the strongest (8-5: accept; 4-1: reject). Spreading the score helps create a distribution for decision-making
5
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The method seems sound, although some more motivations and ablation studies could better convince the reader about specific design choices. Some notations and missing informations make the paper not very easy to read although it is well written.
- Number of papers in your stack
5
- What is the ranking of this paper in your review stack?
2
- Reviewer confidence
Confident but not absolutely certain
- [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
Primary Meta-Review
- Please provide your assessment of this work, taking into account all reviews. Summarize the key strengths and weaknesses of the paper and justify your recommendation. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. In case of an invitation for rebuttal, clarify which points are important to address in the rebuttal.
This paper proposes a method to predict patient-specific organ dose in CT acquisition by combining 2 orthogonal 2D scout CT images, the scan z range in form of a binary 2D image, a tube current modulation map, and 1D patient size profiles. The proposed solution will alleviate the need for a CT scan required for dose personalization in CT imaging, which is of great clinical importance.
Reviewers indicate the novelty and practical value. On the other hand clarity of the method, missing aspects in the ablation study such as the architecture choices and influence of various inputs. Overall interesting approach with a practical value that can benefit from fine-tuning in experiemtns and writing.
- What is the ranking of this paper in your stack? Use a number between 1 (best paper in your stack) and 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).
5
Author Feedback
N/A