List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Weiguo Cao, Benjamin M. Howe, Nicholas G. Rhodes, Sumana Ramanathan, Panagiotis Korfiatis, Kimberly K. Amrami, Robert J. Spinner, Timothy L. Kline
Abstract
Brachial plexopathy is a form of peripheral neuropathy, which occurs when there is damage to the brachial plexus (BP). However, the diagnosis of breast cancer related BP from radiological imaging is still a great challenge. This paper proposes a texture pattern based convolutional neural network, called TPPNet, to carry out abnormal prediction of BP from multiple routine magnetic resonance image (MRI) pulse sequences, i.e. T2, T1, and T1 post-gadolinium contrast administration. Different from classic CNNs, the input of the proposed TPPNet is multiple texture patterns instead of images. This allows for direct integration of radiomic (i.e. texture) features into the classification models. Beyond conventional radiomic features, we also developed a new family of texture patterns, called triple point patterns (TPPs), to extract huge number of texture patterns as representations of BP’ heterogeneity from its MRIs. These texture patterns share the same size and show very stable properties under several geometric transformations. Then, the TPPNet is proposed to carry out the differentiation task of abnormal BP for our study. It has several special characteristics including 1) avoidance of image augmentation, 2) huge number of channels, 3) simple end-to-end architecture, 4) free from the interference of multi-texture-pattern arrangements. Ablation study and comparisons demonstrate that the proposed TPPNet yields outstanding performances with the accuracies of 96.1%, 93.5% and 93.6% over T2, T1 and post-gadolinium sequences which exceed at least 1.3%, 5.3% and 3.4% over state-of-the-art methods for classification of normal vs. abnormal brachial plexus.
Link to paper
DOI: https://doi.org/10.1007/978-3-031-43993-3_46
SharedIt: https://rdcu.be/dnwNR
Link to the code repository
N/A
Link to the dataset(s)
N/A
Reviews
Review #1
- Please describe the contribution of the paper
A novel feature descriptor, triple point pattern (TPP), was proposed to quantitatively represent the texture of volumetric data. Tested on a local dataset of 189 patients for brachial plexus classification,
- 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.
A novel method to extract texture features from 2D and 3D images. A simple neural network which used TPP as a preprocessing module has outperformed the state-of-the-art methods. Experiments are comprehensive, including ablation studies, comparison of different experiment settings, as well as comparison with existing methods. Overall the results support the conclusions.
- 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.
No major weakness was found.
- Please rate the clarity and organization of this paper
Very 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 results are based on the analysis of a local dataset using the proposed method, which is not public. No information was provided in the reproducibility checklist. it is difficult to access the reproducibility of this work.
- 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/2023/en/REVIEWER-GUIDELINES.html
In addition to those typos that are already addressed in the supplementary materials, the authors also need to carefully review the manuscript and correct other typos.
- 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 methodology seems novel; a new dataset with manual segmentations is created, which might enable further studies. Overall the merits weigh over weakness.
- Reviewer confidence
Very 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 #2
- Please describe the contribution of the paper
The authors present a texture-based CNN for brachial plexus classification from T1, T2 and post-gadolinium sequences of MRI. Instead of an image, the network takes as input texture features in the form of triple point pattern matrix with a large number of channels.
- 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 network formulation which takes in triple point pattern matrices as a feature of heterogeneity appears to be novel
- The authors perform several experiments and evaluations using 3 different MRI sequences and 5-fold cross validation.
- 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 methods are a little hard to follow as written and the paper should be corrected for typos/erroneous values.
- The channel size of the matrices going into the network is 325 and the authors use 4 320GB GPUs. High memory requirement seems to be a limitation of this approach. The authors should include some discussion of the limitations of their approach in the conclusion. -The authors do not report standard deviation on the evaluation metrics across the 5 cross-validation folds. They should include this in addition to the average.
- Might be easier to follow the results in the supplemental material if they were reported more concisely/simplified i.e. the confusion matrices could be summarized in a table with average precision/recall across folds.
- 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 authors have not properly filled out the reproducibility form (all NA)
- 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/2023/en/REVIEWER-GUIDELINES.html
See weaknesses
- 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
3
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The paper can be better structured/organized. There are some typos/errors as noted by the authors in the supplemental material that need to be edited in the main manuscript
- 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 presents a texture pattern based convolutional neural network, which uses multiple texture patterns as inputs, for prediction of brachial plexopathy from multisequence MRI. The proposed method is evaluated on a local dataset containing 189 patients (including 141 normal and 41 abnormal) and 462 series.
- 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 exploration of the texture based network in the classification task is original and interesting. In this paper, triple point patterns are proposed to extract texture information. Multiple experiments and comparisons to previous methods are provided. The resulting network obtains better results when compared to the previous work.
- 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.
This was a single institution study, and only focused on BP related to breast cancers limiting the generalizability. Additionally, there is a data imbalance with relatively few abnormal cases.
- 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
No code or data was provided along with the paper.
- 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/2023/en/REVIEWER-GUIDELINES.html
- It would have been valuable to know how the ground truth was established for the abnormal cases.
- Standard deviations or confidence intervals could have been added to the results, to make it easier to assess the value of the method.
- The quality of the current manuscript can be strengthened by the presentation being more succinct in the method and 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?
The authors proposed a novel method, and the reported results over comprehensive experiments show improvement from previous works.
- 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
6
- [Post rebuttal] Please justify your decision
Several comments are addressed in the rebuttal. Although there are concerns regarding the clarity and presentation of the paper, the paper may have some interest for MICCAI.
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 presents a new CNN framework to predict brachial plexopathy from multisequence MRI. The exploration of the texture based network in the classification task is original and valuable. Since the review comments are polarized, I would like to invite rebuttal so that the authors can provide some discussions on those issues raised by Reviewer #2.
Author Feedback
Reviewer #1, Reviewer #2, Review #3: Response: We very much appreciate your comments. All typos and erroneous values have been corrected. The standard deviations have been added in Table 2-5. Reviewer #2: 2.1. The methods are a little hard to follow as written. Response: We apologize your comments. In general, clarity and brevity are essential concerns, particularly with regards to this paper’s layout, which includes an introduction, methodology, experiment, conclusion, and references. The methodology section subdivides into three areas: dataset preparation, triple point pattern (TPP), and TPPNet, while the experiment section covers experimental setup, ablation studies, and comparisons. The structure of our neural network is visually represented in Fig. 2, providing an accessible overview of our methodology. The process involves calculating TPPs and using them to train the TPPNet model. Details about TPPNet’s layer structure, parameters, and filters can be found in Fig. 2. Further specifics, including optimizer, learning rate, batch size, epochs, and gray levels, are discussed in the experiment section. The triple point pattern (TPP), inspired by the directed triangle concept, can be difficult to grasp. To enhance understanding, we have included detailed pattern numbers for both 2D and 3D images in section 2.2, and further demonstrated the construction process in Fig. 1 and Fig. s1. TPP, essentially capturing the frequency of local patterns, is explained using examples and diagrams, especially in relation to 3D images. The demonstration in Fig. s1 aids in understanding TPP’s intricacies. In conclusion, we have meticulously detailed our method’s structure and process, with an emphasis on TPP’s creation and calculation. In our experiment, we conducted ablation studies on several parameters, including gray level image TPP, solo-channel, and multi-channel. Our methodology was compared with five leading approaches using ROC curves and five established evaluation measures, proving its superiority as demonstrated in Tables s7, s8, and s9. 2.2. High memory requirement seems to be a limitation of this approach. Response: We greatly appreciate your comments and concerns regarding GPU and memory usage. Our server houses four Nvidia A100-SXM GPUs, each with 80GB memory, totaling 320GB. For model training, we utilize only one GPU. These details have been clarified in our revised version. Regarding memory usage of deep learning models, our perspective differs. Assuming a gray level of 12 (Comparison Section) and a float32 data type, TPPNet’s input for each case would be around 2.1MB. Comparatively, CNN models in our comparison section require approximately 4MB for each BP image. Hence, our method significantly reduces memory usage. Moreover, our TPPNet uses a lean architecture of only 15 layers, considerably fewer than other models. Due to channel number, our model’s parameter amount evenly matches Inception, but much less shan VGG16 and MobileNet. Thus, we assert that our model holds advantages over renowned CNN models in terms of memory requirements. We will include this information in our updated version. 2.3. The confusion matrices could be summarized in a table with average precision/recall across folds. Response: We value your suggestion. The confusion matrix (CM) is a well-recognized tool for illustrating model performance, often favored by clinicians due to its straightforward representation of error sources. It also conveniently provides foundational metrics such as true positives (TP), false positives (FP), false negatives (FN), and true negatives (TN). In our revised version, we have calculated and summarized all precision/recall values for the three MRI sequences based on these four metrics, doing so in a more streamlined and straightforward manner.
Post-rebuttal Meta-Reviews
Meta-review # 1 (Primary)
- Please provide your assessment of the paper taking all information into account, including rebuttal. Highlight the key strengths and weaknesses of the paper, clarify how you reconciled contrasting review comments and scores, indicate if concerns were successfully addressed in the rebuttal, and provide a clear justification of your decision. If you disagree with some of the (meta)reviewer statements, you can indicate so in your meta-review. Please make sure that the authors, program chairs, and the public can understand the reason for your decision.
The authors have addressed all the concerns from reviewer 2 and considering reviewer 1 and 3 initially favored this paper, I would recommend accept.
Meta-review #2
- Please provide your assessment of the paper taking all information into account, including rebuttal. Highlight the key strengths and weaknesses of the paper, clarify how you reconciled contrasting review comments and scores, indicate if concerns were successfully addressed in the rebuttal, and provide a clear justification of your decision. If you disagree with some of the (meta)reviewer statements, you can indicate so in your meta-review. Please make sure that the authors, program chairs, and the public can understand the reason for your decision.
I recommend acceptance of this paper as the majority of reviewers believe the method and manuscript had merit. The main concerns of R2 were mostly related to writing and structure of the manuscript and not methodological which I think can be fully addressed.
I highly recommend before submission of the final camera ready copy the authors take care to try to clarify the writing.
Meta-review #3
- Please provide your assessment of the paper taking all information into account, including rebuttal. Highlight the key strengths and weaknesses of the paper, clarify how you reconciled contrasting review comments and scores, indicate if concerns were successfully addressed in the rebuttal, and provide a clear justification of your decision. If you disagree with some of the (meta)reviewer statements, you can indicate so in your meta-review. Please make sure that the authors, program chairs, and the public can understand the reason for your decision.
The authors proved a novel method to predict abnormal brachial plexus from MRI. Overall this is a well-written paper and the rebuttal seems adequately addressed reviewers’ comments. Thus I would like to suggest acceptance of the paper. If accepted, the authors may need to change the red font in the table, which is rarely seen in research publications. A bold fond might be a better choice.