List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Yuwen Tan, Xiang Xiang, Yifeng Chen, Hongyi Jing, Shiyang Ye, Chaoran Xue, Hui Xu
Abstract
Delineating and removing brackets on 3D dental models and then reconstructing the tooth surface can enable orthodontists to pre-make retainers for patients. It eliminates the waiting time and avoids the change of tooth position. However, it is time-consuming and labor-intensive to process 3D dental models manually. To automate the entire process, accurate bracket segmentation and tooth surface reconstruction algorithms are of high need. In this paper, we propose a graph-based network named BSegNet for bracket segmentation on 3D dental models. The dynamic dilated neighborhood construction and residual connection in the graph network promote the bracket segmentation performance. Then, we propose a simple yet effective projection-based method to reconstruct the tooth surface. We project the vertices of the hole boundary on the tooth surface onto a 2D plane and then triangulate the projected polygon. We evaluate the performance of BSegNet on the bracket segmentation dataset and the results show the superiority of our proposed method. The framework integrating the segmentation and reconstruction achieves a low reconstruction error and can be used as an effective tool to assist orthodontists in orthodontic treatment.
Link to paper
DOI: https://doi.org/10.1007/978-3-031-43987-2_40
SharedIt: https://rdcu.be/dnwJV
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 proposes a graph-based network named BSegNet for generating a 3D surface model of teeth , from an initial 3D surface model where the teeth contain orthodontic brackets. The method identifies and removes the brackets from the surface model and fills the holes in order to approximate the original teeth.
- 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 work is interesting, a relatively niche domain of dental model processing.
- 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 method appears to be another ad-hoc, specialized network that achieves minor improvements in the case of specialized tooth data.
- 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
Reproducible.
- 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
The authors focus exclusively on graph-based deep learning, with the statement “graph-based network [7, 15, 16, 5, 13] shows its superiority on various tasks on 3D dental models”. However results in Table 1 show that infact, the strongest alternative model seems to be PointMLP, which is not graph-based and not designed for dental models. The authors solution improves only slightly upon the PointMLP, in the case of the authors specialized tooth dataset and task.
Thus, it is not clear why the authors choose to create a specialized model based on graph-based neural networks, rather than PointMLP-based framework.
Table 2: Missing evaluation for Automatic reconstruction using PointMLP[8] and Ours, would be interesting to see.
Timing constraints are mentioned as being important in the abstract and intro, however no timing information is available regarding the authors or other methods.
While the proposed method achieves reasonable average eror, there seem to be certain bracket locations where segmentation has been less effective (Figure 3 d) - why is this?
- 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
4
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Specific data case, relatively minor performance improvement.
- 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
• The authors proposed a graph-based network named BSegNet for bracket segmentation on 3D dental models. • The authors developed a method to reconstruct the tooth surface where brackets are removed. • Empirical evaluation was performed on a real data set, yielding promising 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 studied problem is of high significance. If successful, the resulting tool will have high clinical value. • The proposed methods are technically sound and the results in the empirical study are promising. • The paper is well-structured, clearly written and easy to follow.
- 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 rationale behind the local modules could be better described.
- Please rate the clarity and organization of this paper
Excellent
- 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
• Reproducibility is high, given that the paper is clearly written and code is provided.
- 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
• The rationale behind the local modules could be better described.
- 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
7
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
• See the comments above.
- 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 #3
- Please describe the contribution of the paper
To use 3D modeling to expedite the time in which effective orthodontic treatments are provided (e.g., removal of brackets to retainers creation as a result of digital modeling). In addition to time efficiency, this method also ensures safe keeping in the event a patient loses or breaks their retainers (i.e., digital model will be at hand). The authors propose the graph-based BSegNet for bracket segmentation along with a reconstruction of the tooth surface following bracket removal.
- 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.
Well written, articulate, and interesting paper. Needs for research are clearly defined. The novelty of the paper is in both the proposed method as well as its application (i.e., improving orthodontic treatments through 3D modeling). Results are encouraging.
- 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.
A potential limitation is the dataset. The authors use 80 mesh models, and the dataset may not be too challenging. It would be interesting to see the proposed method applied to more challenging data.
- Please rate the clarity and organization of this paper
Excellent
- 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 only paper out of the ones I have reviewed to provide appropriate code for repeatability and reproducibility. Dataset not available.
- 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
The authors propose a segmentation network that analyzes a 3D object by breaking it down to smaller pieces (mesh cells), understands the connection between the mesh cells using a graph-based approach, and then updates features based on the relationship between the meshes. This allows for a more detailed analysis of the 3D structure. Reconstructing the tooth surface is also interesting and aligns with the authors objectives of using a 3D model for creating retainers with ease. This includes identification of holes in the 3D model, boundary marking, projection, triangulation, etc. Overall, a comprehensive process. Very well written paper that caught my attention from the start. Quantitatively and qualitatively, the results appear to be promising.
- 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
8
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
From the very start this paper caught my attention. Highly interesting application of 3D modeling I would not have even considered. Great paper, very clear, great infographics, great presentation overall and interesting/promising results. Code is also very clear and well written.
- 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
8
- [Post rebuttal] Please justify your decision
My original review of a definite accept still stands. The authors have provided adequate responses/justification to the concerns raised by the reviewers.
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.
The work proposed a graph-based network named BSegNet for bracket segmentation on 3D dental models. Empirical evaluation was performed on a real data set, yielding promising results. It is an interesting paper. Although the reviewers were enthusiastic some concerns were addressed, e.g., inconsistent experimental results, more elaboration on the experimental results, more comparison methods. During the rebuttal period, the authors are encouraged to address these concerns and further improve the paper quality.
Author Feedback
We thank all reviewers for their valuable time and insightful comments. We are encouraged by the positive comments on our proposed application (R1,2,3), novel method (R2,3), promising results (R2,3), high reproducibility (R1,2,3), and clarity (R1,2,3). We appreciate your constructive comments and are happy to address the concerns listed in the Weaknesses.
To R1 Q1. Why do we choose to use a graph neural network as a base network rather than PointMLP? PointMLP is based on PointNet++ and uses residual connections to form a deeper network. PointNet++ uses an encoder-decoder structure and down-samples the points by the farthest point sampling in each stage. However, this hierarchical structure does not show superiority in the dental model segmentation task regarding training speed and segmentation performance. Compared to PointNet++, the graph-based network DGCNN achieves better segmentation results. Therefore, our network is based on DGCNN and aims to design several modules to further improve the performance. The mIoU of BSegNet is much higher (94.60 vs. 93.15) than PointMLP before post-processing which verifies the effectiveness of BSegNet. BSegNet performs better than PointMLP regarding segmentation details.
Q2. Missing the reconstruction evaluation results for automatic reconstruction using PointMLP. The reconstruction evaluation results are as follows. Method / Seg. / Recon. / SD(mm) / MD+(mm) / MD-(mm) / Aver.(mm) (i) Auto / PointMLP/ Ours/ 0.055/ 0.008/ 0.010/ 0.0090 The reconstruction errors of PointMLP for both SD and Average (MD+ and MD-) are much higher than our method. Therefore, dental models processed by BSegNet can achieve low reconstruction errors when using the same reconstruction method.
Q3. No timing information is available. It takes about 20 minutes for an orthodontist to delineate the brackets on a 3D dental model. Our proposed automatic framework needs about 5 seconds to finish the network prediction and the reconstruction process. The post-processing after the network prediction stage takes more time, which is about 20 seconds. Thus, the whole process (including network prediction, post-processing, reconstruction) for one dental model needs about 25 seconds. The proposed automatic framework is much faster than the manual operations on 3D dental models so it can save a lot of time for orthodontists. Since the difference in inference time for different networks is not significant, we have not emphasized it in this paper.
Q4. Certain bracket location segmentation seems to be less effective. In the vast majority of cases, our bracket segmentation algorithm performs excellently. For a few dental models, our bracket segmentation algorithm may not perform perfectly on a few teeth, such as the brackets on the molars (the two innermost teeth) may not be completely removed. Thus, as shown in Fig3(d), the reconstruction results on the molars are relatively concave than the ground truth.
To R2 Q1. The rationale behind the local modules could be better described. We elaborate on the rationale of all the modules in the BSegNet as follows. We update the neighbor information by the distance between features instead of static coordinates information which can better update the neighbor information of points during the training. We use dilated-knn instead of traditional knn as it can get a large receptive field and better encode the local areas. With residual connections, we can stack more blocks to form a slightly deeper network to further promote the capacity. After the post-processing stage which uses the geometric information of the dental models, the prediction results of the neural network could be further refined.
To R3 We are glad you find the proposed application interesting and significant and find our method novel and reproducible.
To MR1 We appreciate your constructive comments. The concerns listed in the weakness are similar to R1,2. Please see our reply to R1,2.
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 been responsive for addressing reviewers’ comments. It significantly improved the paper quality. R3 posted post-rebuttal comments while others not. The AC thinks the strengths overweighed the weakness. It is recommended for publication in MICCAI 2023. The authors should try to incorporate the rebuttal in their camera ready paper.
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.
The experimental section needs the futher improvements.
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 paper received an overall score >6, and the authors addressed most of the concerns raised by the reviewers in the rebuttal.