Abstract

Accurate segmentation of the pulp cavity, root canals, and inferior alveolar nerve (IAN) in dental imaging is essential for effective orthodontic interventions. Despite the availability of numerous Cone Beam Computed Tomography (CBCT) scans annotated for individual dental-anatomical structures, there is a lack of a comprehensive dataset covering all necessary parts. As a result, existing deep learning models have encountered challenges due to the scarcity of comprehensive datasets encompassing all relevant anatomical structures. We present our novel Pulpy3D dataset, specifically curated to address dental-anatomical structures’ segmentation and identification needs. Additionally, we noticed that many current deep learning methods in dental imaging prefer 2D segmentation, missing out on the benefits of 3D segmentation. Our study suggests a UNet-based approach capable of segmenting dental structures using 3D volume segmentation, providing a better understanding of spatial relationships and more precise dental anatomy representation. Pulpy3D contributed in creating the seeding model from 150 scans, which helped complete the remainder of the dataset. Other modifications in the architecture, such as using separate networks, one semantic network, and a multi-task network, were highlighted in the model description to show how versatile the Pulpy3D dataset is and how different models, architectures, and tasks can run on the dataset. Additionally, we stress the lack of attention to pulp segmentation tasks in existing studies, underlining the need for specialized methods in this area. The code and Pulpy3D links can be found at https://github.com/mahmoudgamal0/Pulpy3D

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/1419_paper.pdf

SharedIt Link: pending

SpringerLink (DOI): pending

Supplementary Material: N/A

Link to the Code Repository

https://github.com/mahmoudgamal0/Pulpy3D

Link to the Dataset(s)

https://drive.google.com/drive/folders/1M5iU1urLOp1rSxKOm7WCzodAKcZrqT5O?usp=sharing

BibTex

@InProceedings{Gam_Automatic_MICCAI2024,
        author = { Gamal, Mahmoud and Baraka, Marwa and Torki, Marwan},
        title = { { Automatic Mandibular Semantic Segmentation of Teeth Pulp Cavity and Root Canals, and Inferior Alveolar Nerve on Pulpy3D Dataset } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
        year = {2024},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15008},
        month = {October},
        page = {pending}
}


Reviews

Review #1

  • Please describe the contribution of the paper

    The study introduces the Pulpy3D dataset, specifically curated to address the scarcity of comprehensive datasets covering all necessary dental-anatomical structures, enabling accurate segmentation and identification of the pulp cavity, root canals, and inferior alveolar nerve (IAN) crucial for orthodontic interventions. Additionally, the study proposes a UNet-based 3D segmentation approach, which enhances spatial understanding and provides more precise dental anatomy representation, filling the gap left by existing deep learning methods that predominantly rely on 2D segmentation in dental imaging.

  • 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.

    Additional clinically relevant annotations generated for an existing dental dataset.

  • 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.

    Limited technical novelty with only marginal gains with the addition of gated-attention to prior state-of-the-art model. Insufficient details on clinical validation.

  • 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

    Well motivated study to provide additional clicinally relevant annoations for the field. However, insufficient technical novelty and lacks details of the clinical validation of the labels.

  • 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

    Reject — should be rejected, independent of rebuttal (2)

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    Insufficient technical novelty and lacks details of the clinical validation of the labels.

  • 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



Review #2

  • Please describe the contribution of the paper

    The paper proposes a novel Pulpy3D dataset for dental anatomical structure segmentation and identification needs. It suggests a UNet-based 3D segmentation network and validates its effectiveness on the proposed dataset.

  • 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.
    • Propose a new Pulpy3D dataset of 443 scans, with annotated pulp segmentation labels, pulp instance segmentation labels, and the combined labels for merging the pulp with IAN.
    • Detailed descriptions of data annotation and collection are provided.
  • 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.
    • More details are encouraged to indicate annotation quality.
    • Several papers about pulp segmentation are not discussed and cited in this paper, some of them are based on 3D segmentation, not 2D segmentation on slices. Including but not limited to: [1] Zheng, Zhiyang, et al. “Anatomically constrained deep learning for automating dental CBCT segmentation and lesion detection.” IEEE Transactions on Automation Science and Engineering 18.2 (2020): 603-614. [2] Duan, Wei, et al. “Refined tooth and pulp segmentation using U-Net in CBCT image.” Dentomaxillofacial Radiology 50.6 (2021): 20200251. [3] Tan, Minhui, et al. “Dental Anatomy Segmentation from Cone Beam CT Images.” 2023 IEEE 20th International Symposium on Biomedical Imaging (ISBI). IEEE, 2023. [4] Jiang, Benxiang, et al. “Dental pulp segmentation from cone-beam computed tomography images.” The Fourth International Symposium on Image Computing and Digital Medicine. 2020.
    • Experiments are mainly conducted for IAN segmentation. Lack of experiments specifically for pulp segmentation.
  • 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

    See above

  • 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?

    This paper provides a novel dataset for dental anatomical structure segmentation and identification needs, but more discussions and experiments are encouraged.

  • Reviewer confidence

    Somewhat confident (2)

  • [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

    Thanks to the rebuttal. The clarifications are very helpful.

    From what I now understand, the main contributions of this paper are: 1) presenting a novel Pulp3D dataset with 3D anatomy annotations in CBCT images (can serve as a benchmark), and 2) providing an efficient seeding model that eases future manual annotation of new data (validated using 150 cases from the dataset). The authors claim that the dataset and the seeding model will be released.

    However, the original manuscript does not effectively point out these contributions, making it very difficult for readers to understand the main purpose of this paper. I strongly recommend that the authors incorporate these explanations more clearly into the manuscript, and consider reorganizing the paper to align with these contributions, especially in the abstract and Sec 3. In the current manuscript, Sec 3 lacks clear motivation for the models and the objectives of the experiments (why single-task and multi-task experiments are designed and conducted). It’s very crucial to first clarify the goal of these experiments (to identify the best seeding model) and the rationale behind the experimental design. Additionally, I encourage the authors to consider expanding the training dataset for the seeding models (currently only 150 out of 443 cases), to potentially enhance model robustness and validity.

    Minor: I recommend revising Fig 2 to reduce redundancy and save some space. The primary difference seems only to be one v.s. two decoders. You could initially illustrate the two pipelines only using high-level structures in the first row (input->components->output), and detail the encoder and decoder structures in the second row.



Review #3

  • Please describe the contribution of the paper

    A dataset named Pulpy3D has been presented to address dental-anatomical structures’ segmentation and identification needs. Automatic segmentation of the proposed dataset has also been performed to suggest that Unet-based 3D segmentation can provide a better understanding of space and more precise dental anatomy representation.

  • 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.
    1. The proposed 3D dental-anatomical segmentation dataset is novel
    2. Well-written Methods and Results sections
    3. Noticeable performance boost
  • 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.
    1. The contribution statement in the abstract should be clearer
    2. Runtime has not been reported
    3. The rationales of using the incorporated network and the associated training parameters are not discussed.
  • 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. Abstract: Please enhance the clarity of your statement on the previous works’ shortcomings and your contribution(s) to resolve them.
    2. Abstract: No quantitative result has been presented in the Abstract.
    3. Introduction: Very limited citations to back up the authors’ statements.
    4. Network architecture: Please discuss the rationale of using the incorporated network architecture.
    5. Fig. 2: The annotations are hard to read. Please consider improving the readability.
    6. How did you optimize different training parameters?
    7. Please report your runtime(s).
  • 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?

    Factors:

    1. Novelty of the dataset
    2. Performance boost
  • 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

    Accept — should be accepted, independent of rebuttal (5)

  • [Post rebuttal] Please justify your decision

    My original decision was also ‘accept’. In addition, I am satisfied with the authors’ response to the reviewers’ comments.




Author Feedback

We are thankful for the feedback. We provide our response in 5 main points.

1- Main Contribution One comment is that our contribution suffers from limited technical novelty with only marginal gains. This is not true as we pointed out in the paper motivation how the literature lacks a comprehensive dataset for pulp segmentation. We presented Pulpy3D, a unique and novel contribution, that will encourage dental research and new deep-learning algorithms. Our contribution is not using a prior state-of-the-art model and just adding gated attention to it. We clearly stated in the data collection section and showed in the Baseline comparison how our change aided in creating the seeding model from 150 scans which helped complete the remaining of the dataset. Therefore, the use of gated attention was backed up by its results on the validity of the seeding model. Other modifications in the architecture like using separate networks, one semantic network, and a multi-task network were highlighted in the model description to show how versatile the Pulpy3D dataset is and how different models, architectures, and tasks can run on the dataset. So, to say that the contributions are marginal and summarized by just adding an attention unit is underestimating our contribution.

  1. Annotations As for clinical validation, we had a team of dentists who were led by an expert with 15 years of experience. The team manually annotated the pulp cavity and root canals using the ITK-Snap tool. We can quote from 24 references similar statements where they stated the number of scans and the number of years of experience with their annotators. Since many of these datasets were not published, we can’t validate their claims. However, our dataset will be publicly available leading to open-source contribution and intrinsic validation from the community. However, we added more details In the pre-annotation phase, we had a group of junior annotators who went under a special training program led by our expert. For the best two of them, inter-examiner and intra-examiner scores are calculated to segment 6 scans, 3 weeks apart. The intraclass correlation coefficient (ICC) for the inter-examiner and intra-examiner reliability for volumetric pulpal assessments were (ICC = 0.986, 95% CI = 0.937, 0.997) and (ICC = 0.971, 95% CI = (0.933, 0.981), respectively.

3- Literature Review In the literature review, we focused on the segmentation task of CBCT, particularly the use of 2D and 3D segmentations. One review commented that we did not cite some important papers that dealt with 3D segmentations. The review mentioned 4 refs. We already cited [1,2] of the 4 refs as refs [24, 5 ]in our paper. We missed the refs [3, 4] but we will add to the final version. As mentioned in the related work and dataset motivation [1, 2] work on 2D slices and used 2D-Conv rather than volumes and using 3D-Conv. The interpreted 3D arises from stacking multiple 2D slices which is not an effective 3D representation as it lacks the volumetric features in a pure 3D volume.

4- Experiments One of the comments is that our experiments were conducted for IAN segmentation and not enough for pulp segmentation. This is inaccurate. As explained in the Model and Experiments sections, we created a special seeding model for pulp segmentation only. We performed experiments using single networks for both pulp and IAN. We trained a single semantic network and a multi-task two-decoder network for both pulp and IAN with distinct labels for each. Training parameters and runtime environment were indicated in mentioned sections in addition to complete configuration files in the code which will be available once accepted.

5- Running time We thank the reviewers for asking about run time. We calculated the inference time of the scans. The mean time in (seconds) is as follows: Separate Network: 4.3 sec. Less than 5 seconds. One semantic network: 3.84 sec. The common network: 5.57 sec.




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.

    Reject

  • Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’

    The primary contribution of this paper is the annotation of the dataset; however, the technical novelty is limited, as existing 3D segmentation networks, such as nnUNet, are likely to achieve excellent results on this task. Moreover, the public dataset only contains partial upper teeth scans, making it an insufficient resource for comprehensive pulp cavity segmentation research.

  • 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).

    The primary contribution of this paper is the annotation of the dataset; however, the technical novelty is limited, as existing 3D segmentation networks, such as nnUNet, are likely to achieve excellent results on this task. Moreover, the public dataset only contains partial upper teeth scans, making it an insufficient resource for comprehensive pulp cavity segmentation research.



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’

    Two reviewers recommend accept and one reject. The rebuttal well addresses comments from the latter reviewer as well as the other reviewers. With the suggested edits by the reviewers integrated in the manuscript, I recommend acceptance.

  • 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).

    Two reviewers recommend accept and one reject. The rebuttal well addresses comments from the latter reviewer as well as the other reviewers. With the suggested edits by the reviewers integrated in the manuscript, I recommend acceptance.



Meta-review #3

  • 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’

    As pointed out by the reviewers, the main contribution of this work is a new dental CBCT dataset, which the authors promised to release to the public once the paper is accepted. The technical novelty is relatively weak. Considering the pros and cons of this work, I lean towards accepting it.

  • 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).

    As pointed out by the reviewers, the main contribution of this work is a new dental CBCT dataset, which the authors promised to release to the public once the paper is accepted. The technical novelty is relatively weak. Considering the pros and cons of this work, I lean towards accepting it.



back to top