List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Haoran Dou, Luyi Han, Yushuang He, Jun Xu, Nishant Ravikumar, Ritse Mann, Alejandro F. Frangi, Pew-Thian Yap, Yunzhi Huang
Abstract
Tumor infiltration of the recurrent laryngeal nerve (RLN) is a contraindication for robotic thyroidectomy and can be difficult to detect via standard laryngoscopy. Ultrasound (US) is a viable alternative for RLN detection due to its safety and ability to provide real-time feedback. However, the tininess of the RLN, with a diameter typically less than 3 mm, poses significant challenges to the accurate localization of the RLN. In this work, we propose a knowledge-driven framework for RLN localization, mimicking the standard approach surgeons take to identify the RLN according to its surrounding organs. We construct a prior anatomical model based on the inherent relative spatial relationships between organs. Through Bayesian shape alignment (BSA), we obtain the candidate coordinates of the center of a region of interest (ROI) that encloses the RLN. The ROI allows a decreased field of view for determining the refined centroid of the RLN using a dual-path identification network, based on multi-scale semantic information. Experimental results indicate that the proposed method achieves superior hit rates and substantially smaller distance errors compared with state-of-the-art methods.
Link to paper
DOI: https://link.springer.com/chapter/10.1007/978-3-031-16440-8_25
SharedIt: https://rdcu.be/cVRvQ
Link to the code repository
https://github.com/wulalago/RLNLocalization
Link to the dataset(s)
N/A
Reviews
Review #1
- Please describe the contribution of the paper
This paper presented a learning-based framework to identify the RLN from a US image for pre-operative assessment of contraindication for robotic thyroidectomy, which introduces Bayesian shape alignment and Locate-Net for geometrical constraints and localization result refinement, respectively.
- 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.
This paper is well-written and organized. The idea of mimicking the standard approach surgeons take to identify the RLN according to its surrounding organs and using the inherent relative spatial relationships between organs is interesting and proved helpful.
- 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 about the experiments and analysis related to the results should be included.
- 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
I believe that the obtained results can be reproduced.
- 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
It would be better to explain more about the relationships among the surrounding organs.
- 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 novelty of the proposed method and the experiments are sufficient for weak accept.
- 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 #3
- Please describe the contribution of the paper
Clinically, this paper helps clinicians to find the recurrent laryngeal nerve with ultrasound images. Technically, it puts forward a coarse-to-fine method to find the nerves in a segmentation network.
- 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.
This paper puts forward a method which could solve a practical clinical problem. The method’s performance is good. The study design is impressive: lots of patients were involved; the new method is compared with many current methods; an ablation study shows that both local information and global context contributed to the good performance; visualization of the result is convincing.
- 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 would be better if the dataset could include negative cases, so that the accuracy is more meaningful.
For ultrasound images, it is better if authors could discuss how to select qualified frames from all images, because each sonographer could have a different acquisition habit. And this is a protocol that could be used for a multi-center study.
- 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 will not be public: this is a concern. However, authors mention that the network and trained weights could be found later.
- 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
It would be better if the dataset could include negative cases, so that the accuracy is more meaningful.
For ultrasound images, it is better if authors could discuss how to select qualified frames from all images, because each sonographer could have a different acquisition habit. And this is a protocol that could be used for a multi-center study.
- 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?
This paper puts forward a method which could solve a practical clinical problem. The method’s performance is good. The study design is impressive: lots of patients were involved; the new method is compared with many current methods; an ablation study shows that both local information and global context contributed to the good performance; visualization of the result is convincing.
- Number of papers in your stack
3
- What is the ranking of this paper in your review stack?
1
- 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 #4
- Please describe the contribution of the paper
The authThis paper proposed a recurrent laryngeal nerve (RLN) localization method in ultrasound images. The Bayesian shape alignment is applied to obtain the ROI center according to the surrounding organ segmentation results. Small and large ROI patches are cropped from the ultrasound image and then fed into a coordinates regression network.
- 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 authors proposed a novel approach to obtaining the candidate ROI centers that utilized prior clinical knowledge. In particular, Bayesian shape alignment introduces the prior knowledge that models the spatial relationship of RLN, common carotid arteries (CCA), thyroid, and trachea, which gives a good initial location for subsequence refinement. The final results of RLN localization are significantly better than the baseline heatmap-based/regression-based methods.
- 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 writing is basically good, however, many errors are not checked in the manuscript, e.g., the result of heatmap-based method UNET is put in the coord-based section in Table. 1, missing the first name of the first author in reference [1], etc. Some details of the dataset are missing, the total scan frame number should be given, the original resolution of the frames is missing, and how the disagreement of different ground truths is handled as three clinical experts annotated the data. The Tabel.2 shows that the distance/hit rate are 7.52 pix/89.0% for left RLN, which is already significantly better than the baseline methods. Why the Coord-based/heatmap-based methods’ performance is so terrible? Could the authors interpret the poor results of the baseline models? Were these models well trained? Many hyperparameters are set without explanation, such as the crop size of local/global patches, the learning rate and batch size of model training, and the threshold distance for hit rate evaluation. If you follow the setting of previous works, please cite them.
- Please rate the clarity and organization of this paper
Poor
- 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
All the experiments are conducted with private dataset. The performance of baseline method is too low, I don’t know if the authors well trained the baseline models, however, the authors didn’t give any explanation, such as the training losses curves of baseline models.
- 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
Besides the above comments, I have some further suggestions. The authors can test their method in some other public localization datasets to improve reproducibility. There would be more structure priors in 3D images, the author can consider extending the work in 3D images.
- 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?
The proposed method is reasonable and novel. However I am not convinced by the results, especially the baseline models’ performance.
- Number of papers in your stack
1
- 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
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.
Overall, the reviewers recognized the value of the paper.
- 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).
2
Author Feedback
Reviewer #1 The relationships among the surrounding organs. Answer: Thanks for your comments! Despite the individual difference, the anatomical organ-level relationships between the recurrent laryngeal nerve (RLN) and its surroundings are fixed at the cohort level. Anatomically, RLN branches from the vagus nerve, bypasses the carotid common artery (CCA), runs upward along the esophagus-tracheal groove and enters the larynx through the back of the cricothyroid joint.
Reviewer #3 Include negative cases, so that the accuracy is more meaningful. Discuss how to select qualified frames from all images.
Answer: Thank you for your comments! We will include the negative cases (without RLN in the US images) in the future study. Each selection passed the inter-and intra-consistency test, ensuring that the selection achieves high consistency across the three involved experts and has approximate annotations given by an identical physician in a week interval. All the acquisitions must contain a clear RLN and encapsulate the adjacent organs, e.g., esophagus trachea, as complete as possible. At present, there is no gold standard to collect the cross-section of the RLN, and we would like to formulate a standard collection protocol in the future.
Reviewer #4
The writing is basically good, however, many errors are not checked in the manuscript Answer: We will revise all these errors in the camera-ready version.
Some details of the dataset are missing, the total scan frame number should be given, the original resolution/on of the frames are missing, and how the disagreement of different ground truths is handled as three clinical experts annotated the data. Answer: We will detail our collection protocol during the acquisition in the camera-ready version. Each selection passed the inter-and intra-consistency test, ensuring that the selection achieves high consistency across the three involved experts and has approximate annotations given by an identical physician in a week interval.
The Tabel.2 shows that the distinct rate is 7.52 pix/89.0% for left RLN, which is already significantly better than the baseline methods. Answer: This also validates the effectiveness of the prior-knowledge-based method. RLN is indeed tiny in size as compared to its surrounding organs and background. A single network, e.g., the mentioned baselines, is difficult to complete the accurate detection of RLN and is easy to result in the strayed centroids of RLN. As compared to the baselines, our method searched the candidate centroids of RLN within the plausible region depending on the shape-prior constructed by the surrounding organs, eliminating the off-target errors.
The performance of the baseline method is too low Answer: Although the baseline comparisons listed in Table.1 are classical detection neural networks, they are not specifically designed for tiny object detection. Especially, features of the tiny target are easy to be merged in these methods, making it difficult to detect the tiny RLN accurately and is easy to result in the strayed centroids of RLN. When training the baseline comparisons, we chose the optimal hyper-parameters for each baseline method based on the grid search strategy.
Many hyperparameters are set without explanation, such as the crop size of local/global patches, the learning rate and batch size of model training, and the threshold distance for hit rate evaluation. Answer: The crop size of local/global patches was based on the alignment bias (+-12 pixel). And the threshold distance in the Hit Rate was set according to the largest radius of RLN in the collected data.