List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Yuxin Zou, Haoran Dou, Yuhao Huang, Xin Yang, Jikuan Qian, Chaojiong Zhen, Xiaodan Ji, Nishant Ravikumar, Guoqiang Chen, Weijun Huang, Alejandro F. Frangi, Dong Ni
Abstract
Standard plane (SP) localization is essential in routine clinical ultrasound (US) diagnosis. Compared to 2D US, 3D US can acquire multiple view planes in one scan and provide complete anatomy with the addition of coronal plane. However, manually navigating SPs in 3D US is laborious and biased due to the orientation variability and huge search space. In this study, we introduce a novel reinforcement learning (RL) framework for automatic SP localization in 3D US. Our contribution is three-fold. First, we formulate SP localization in 3D US as a tangent point-based problem in RL to restructure the action space and significantly reduce the search space. Second, we design an auxiliary task learning strategy to enhance the model’s ability to recognize subtle differences crossing Non-SPs and SPs in plane search. Finally, we propose a spatial-anatomical reward to effectively guide learning trajectories by exploiting spatial and anatomical information simultaneously. We explore the efficacy of our approach on localizing four SPs on uterus and fetal brain datasets. The experiments indicate that our approach achieves a high localization accuracy as well as robust performance.
Link to paper
DOI: https://link.springer.com/chapter/10.1007/978-3-031-16440-8_29
SharedIt: https://rdcu.be/cVRvU
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 an RL based approach to search for the standard plane in 3D ultrasound. Contributions/claims:
- This paper introduces a tangent-point based Standard Plane (SP) localization method that restricts the search space making optimization easier. Instead of relying on pre-registration it uses imitation learning to initialize the agent, thereby making optimization further easier.
- It proposes an auxiliary task to enable differentiation between non-SPs and SPs.
- It proposes a reward function that uses spatial and anatomical information to guide the agent.
- 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.
Strength:
- The paper is very well written
- The idea of using a tangent point to uniquely represent a SP is pretty nifty as the search space is reduced to the surface of the sphere of a given radius.
- The paper clearly formulates the problem as a Reinforcement Learning Problem by clarifying the state and action space and reward function
- The figures are clear and well-produced.
- The supplementary training video is quite cool!
- 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 agent is pre-trained by supervised learning with the ground truth data. The auxiliary task learns a representation of the SP close to the target SP by supervised learning. Although they are great tools, they raise doubt to what degree the success of the agent depends on the RL algorithm, but not due to the supervised learning segments. That also poses the question “do we even need RL here?” An ablation study might be useful to answer these questions. The results section compares the method with a regression based and registration based method, but it does not say anything about what would happen if we only use the imitation learning module along with the auxiliary task to refine the features.
-
How is a tangent-point based solution better than the directional cosine and distance from the origin based translation? Fig. 2 makes some effort at explaining it through some great visualizations, but the benefits are still not clear in Section 2.1. “The coupling among the directions makes actions dependent”- It’s not clear since x, y and z are orthogonal and thus have no impact on each other.
-
- 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 paper mentions of the training process some relevant hyper-parameters in section 2.3 and 3.1. But the submission did not accompany any code. In the absence of a codebase, reproduction might be challenging.
- 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
Some clarifications that might improve readability:
- The actions are defined as change in the tangent point location along the x, y and z axis. The section after that definition is confusing: “we model the agent-environment interaction as a multi-stage motion process by progressively scaling down the step size from 1.0 to 0.01 when the agent appears to oscillate for three steps”. So, what is the final definition of the actions?
- Explain how the 2D image is reconstructed from the tangent point.
- It’s not clear how the ground truth heat-maps were computed. Were those available with the dataset? If not, briefly describe the heat-map generation process.
- Improve the Fig. 1 caption. Currently it says nothing about the figure. Figures should be self-sufficient.
- 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?
Well written paper, clear description of the methodology.
- 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 #2
- Please describe the contribution of the paper
The paper defines a new tangent-point-based plane formulation in RL to restructure the action space and significantly reduce the search space; designs an auxiliary task learning strategy to enhance the model’s ability to recognize subtle differences crossing Non-SPs and SPs in plane search; and proposes a spatial-anatomical reward to effectively guide learning trajectories by exploiting spatial and anatomical information simultaneously.
- 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 formulation of the SP localization is essential for optimizing the RL framework. This work proposes a new formulation for SP localization in RL. The proposed formulation is unrestricted by directional cosines coupling with less action space than the previous ones (6 < 8). The reward function affect the optimal searching policy with the action. Considering the abnormal data, the reward function encourage the agent to perceive anatomical information.
- 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 anatomical structure reward is essentially a reward based on landmarks reward. For some designed module, there are lack of ablation study to demonstrate the significance of the tangent point formulation and imitation learning. It only show the impact of SCSP and SAR. Compared with model which without SCSP and SAR, the performance of proposed model has no significant improvements especially in SSIM, in ablation studies. .
- 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 model is described in sufficient detail, and the authors outline their experimental procedure clearly. The supplementary provides a video which shows the navigation SPs in 3D US.
- 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
The paper is well-written and proposes a solution to a challenging clinical problem.They demonstrate significant improvements over the state-of-the-art methods. Their model has robust performance on abnormal data. In ablation studies, the performance of proposed model has no significant improvements especially in SSIM. The initialization of agent is important to learning performance. The proposed model adopt imitation learning as an initialization of the agent. There is also lack of ablation study to demonstrate the efficiency of the imitation learning. In the comparison experiments, the proposed model compare with the RL_wsadt and RL_avp which are based on the traditional formulation (8 parameters). I wonder the performance of the proposed model based on traditional formulation with SCSP and SAR module.
- 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
5
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The formulation of the SP localization is essential for optimizing the RL framework. This work proposes a new formulation for SP localization in RL. The proposed formulation is unrestricted by directional cosines coupling with less action space than the previous ones (6 < 8). The reward function encourage the agent to perceive anatomical information. The tangent point formulation has been applied in other image application such as 3d reconstruction of fetal brain in MRI. In results of comparison experiments, the proposed method outperforms all of the others on the most of the metrics. For the ablation studies, performance of proposed model has no significant improvements especially in SSIM. but some designed module lack ablation studies.
- Number of papers in your stack
5
- What is the ranking of this paper in your review stack?
3
- 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
The authors propose a localisation framework that uses RL with a novel tangent-based formulation and a novel anatomical landmark-based reward. Additionally imitation learning-based initialisation and an informative auxiliary task of state-content-similarity prediction, are used to aid learning.
- 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 way to frame the problem of localisation with a novel tangent-based formulation and a novel anatomical landmark-based reward; combining an auxiliary task with RL problems is also fairly uncommon and thus has some novelty
- Sufficient details are provided in order for the paper to be reproducible
- Ablation studies and experiments comparing different methods are useful and insightful
- 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.
- Proper justification is not provided for some claims e.g. “rely on initial registration to ensure data orientation consistency. They are easily trapped when pre-registration fails”; evidence in the form of citations or preliminary experiments would be useful in order to justify claims like this
- Some unclear wordings should be clarified e.g. “ability to recognize subtle differences crossing Non-SPs and SPs in plane search”; what does ‘crossing’ mean here? how do you define subtle (maybe provide an example of what is subtle)? etc.
- The components of the formulation e.g. imitation learning-based initialisation, anatomical feature-based reward and tangent-based formulation are mostly well justified but would have been good to see some more citations of previous works that utilise some of these techniques
- Statistical tests to compare methods are not performed; these help to justify claims of superiority of one method over another, without them it is difficult to draw meaningful conclusions from the results; however, spread is reported which aids in drawing some conclusions
- 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
- Sufficient details are provided in order for the paper to be 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/2022/en/REVIEWER-GUIDELINES.html
Overall this is a strong contribution to the field with only minor weaknesses, which if addressed, would make the paper stringer. In particular if some of the comments about justifying some claims and performing statistical tests, that would make the manuscript even stronger and more self-contained.
- 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-organised, presents a novel methodology and mostly justifies the methods well, whether it be in the design choice stage or post-results; some small weaknesses are present (see above), which if addressed, would make the manuscript stronger.
- Number of papers in your stack
4
- 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.
The reviewers agree that this paper should be accepted. Please take the reviewer’s comments into account for the camera ready version!
- 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).
5
Author Feedback
N/A