List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Yanzhen Liu, Sutuke Yibulayimu, Yudi Sang, Gang Zhu, Yu Wang, Chunpeng Zhao, Xinbao Wu
Abstract
Pelvic fracture is a severe type of high-energy injury. Segmentation of pelvic fractures from 3D CT images is fundamental for trauma diagnosis, evaluation, and treatment planning. Manual delineation of the fracture surface can be done in a slice-by-slice fashion but is slow and error-prone. Automatic fracture segmentation is challenged by the complex structure of pelvic bones and the large variations in fracture types and shapes. This study proposes a deep-learning method for automatic pelvic fracture segmentation. Our approach consists of two consecutive networks. The anatomical segmentation network extracts left and right ilia and sacrum from CT scans. Then, the fracture segmentation network further isolates the fragments in each masked bone region. We design and integrate a distance-weighted loss into a 3D U-net to improve accuracy near the fracture site. In addition, multi-scale deep supervision and a smooth transition strategy are used to facilitate training. We built a dataset containing 100 CT scans with fractured pelvis and manually annotated the fractures. A five-fold cross-validation experiment shows that our method outperformed max-flow segmentation and network without distance weighting, achieving a global Dice of 99.38%, a local Dice of 93.79%, and an Hausdorff distance of 17.12 mm. We have made our dataset and source code publicly available and expect them to facilitate further pelvic research, especially reduction planning.
Link to paper
DOI: https://doi.org/10.1007/978-3-031-43996-4_30
SharedIt: https://rdcu.be/dnwO4
Link to the code repository
https://github.com/YzzLiu/FracSegNet
Link to the dataset(s)
https://github.com/YzzLiu/FracSegNet
Reviews
Review #2
- Please describe the contribution of the paper
The manuscript proposes a deep learning based method for pelvic fracture segmentation, a relatively unexplored application that is highly relevant. A distance-weighted loss refines the accuracy of the segmentation of fragments near the fracture site. A 5-fold validation experimnet on 100 CT scans with fractured pelvises shows that the proposed method outperforms max-flow segmentation, and an ablation study shows the value of the distance weighted loss.
- 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 application of fracture segmentation is highly relevant and valuable for related tasks.
- The achieved performance near the fracture site of 93.79% DICE in cross-validation is significant.
- The experiments validate the advantage of learning-based approaches generally as well as the specific distance-weighted loss.
- 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.
- It is not immediately clear whether the segmentation identifies the fragments or whether the exposed surfaces created by the fractures are localized.
- 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
Code will be made available, and the CTPelvic1K dataset is public. I encourage the authors to make their annotations of fractures 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
My main concern is that it is not immediately clear whether the exposed 2D surfaces of the fractures are localized or if the individual fragments are segmented in 3D. Both are valuable tasks, and they are well-suited to accomplishing the clinical goal, which is fracture reduction through fragment manipulation. Some discussion of the difference between the two may be appropriate.
Overall, it should be emphasized this is strong 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
7
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Fracture segmentation in CT images is highly relevant, with downstream applications including fragment positioning as mentioned in the discussion. This capability is a critical step toward automonomous pelvic fixation, in which fragments must be identified and manipulated. The results show that fragments and fractures are adequately segmented, and the novel distance-weighted loss is shown to be advantageous. Thus the paper is well motivated, well written, and shows good results.
- 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 #1
- Please describe the contribution of the paper
The paper describes a bone fragment segmentation for pelvic fractures. For that a deeply supervised segmentation network is utilized with a new point to fracture distance weighted cost function. The dataset is planned to be published.
- 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.
Overall it is an axcellent paper with little room for criticism: Introduction of a new topic related cost function with sound motivation Clear presentation of the topic Extensive comparison to a large number of other algorithms Good discussion of the results
- 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 small weakness of the paper is, that an evaluation of the number of found particles in relationship to size and/or distance of particle to next part is missing. That could help to understand the limits of the work.
- 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
Paper is reproducable.
- 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
If you have still space the above mentioned evaluation would be nice.
- 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?
The paper is clearly written, the topic is relevant, the presented method is new, the evaluation could be a little bit more elaborated especially on the task of fracture particle delineation.
- 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
The paper introduces a pipeline for pelvic fracture segmentation from 3D CT images. The pipeline consists of two U-Net-like models. The first segments the anatomy of the pelvis and the second segments the fracture. In the fracture segmentation model, the authors design a new loss function that weights each pixel according to its distance to fractures. Then, auxiliary losses are added to each layer in the decoder for multi-scale deep supervision. Moreover, to better stabilize the training, the smooth transition is proposed to weigh the auxiliary losses in the function of the training iteration.
- 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 methods in the pipeline are sound. The authors have identified the difficulty of the automatic pelvic fracture segmentation, i.e. the important variance of the fracture such as intensity, geometry and position. Hence, the authors propose the distance-weighted Dice loss and cross-entropy loss. Furthermore, to avoid ignoring global information, the auxiliary losses that are weighted by the number of training iteration balance between decoder layers the global and local attention.
- The application is novel. The authors state that the work may be the first automatic pelvic fracture segmentation pipeline.
- The paper is well-written and pleasant to read.
- The evaluation is well organised and the authors have published their code online.
- 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 proposed methods bring with them a considerable number of hyper-parameters. They could be sensible to datasets.
- 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 dataset is in-house due to the unavailability of relevant annotations. The code has been published online. The employed hyper-parameters are given.
- 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
- A discussion that justifies the chosen values of hyperparameters is welcome. Moreover, for perspective, the study of their robustness on different datasets can be valuable if further annotations will be ready.
- 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?
The paper is well-written and the application is novel. Moreover, the application is of great value and the evaluation is promising.
- 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
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.
All reviewers agree that this work is worthy of acceptance, as it describes a method for a new problem, includes convincing validation, and is very well presented.
Reviewers raised minor concerns and clarifying questions (segmentation of fractures being direct or indirect; analysis of failure cases w.r.t. fragment size) that should be addressed in an updated version.
Author Feedback
We appreciate the positive feedbacks and constructive comments from the reviewers. We have modified the manuscript accordingly and would like to make the following clarifications.
Reviewer #1: In order to define a consistent labeling strategy in annotation, we restrict the number of fragments of each bone to three. Therefore, the small bone fragments are not separately labeled in the dataset. Currently, they can be identified by post-processing when they are isolated from other fragments. We consider this as a minor limitation and will assess its influence in future study.
Reviewer #2: Our method directly deals with bone fragment segmentation instead of localizing the fracture surfaces. Note that the contact fracture surface is the part where the fragments collide and overlap due to compression and impact and does not include completely isolated parts. The contact surface is used to assign spatial weights in the training loss. We have incorporated the clarifications into the manuscript to improve readability.
Reviewer #3: Our implementation is based on nnU-net. Therefore, most of the network hyper-parameters and data pre-processing are self-configured. The hyper-parameters involved in the deep supervision and smooth transition are determined empirically. Due to the space limit, we plan to perform a sensitivity test in future study. We plan to report the data distribution in detail upon the opening of the dataset.