List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Chenchu Xu, Dong Zhang, Yuhui Song, Leonardo Kayat Bittencourt, Sree Harsha Tirumani, Shuo Li
Abstract
Contrast-free liver tumor detection technology has a significant impact on clinics due to its ability to eliminate contrast agents (CAs) administration in the current tumor diagnosis. In this paper, we proposed a novel ternary knowledge transferred teacher-student DRL (Ts-DRL) as a safe, speedy, and inexpensive contrast-free technology for liver tumor detection. Ts-DRL leverages a teacher network to learn tumor knowledge after CAs administration, and create a pipeline to transfer teacher’s knowledge to guide a student network learning of tumor without CAs, thereby realizing contrast-free liver tumor detection. Importantly, Ts-DRL possesses a new ternary knowledge set (actions, rewards, and features of driven actions), which for the first time, allows the teacher network to not only inform the student network what to do, but also teach the student network why to do. Moreover, Ts-DRL possesses a novel progressive hierarchy transferring strategy to progressively adjust the knowledge rationing between teachers and students during training to couple with knowledge smoothly and effectively transferring. Evaluation on 325 patients including different types of tumors from two MR scanners, Ts-DRL significantly improves performance (Dice by at least 7%) when comparing the five most recent state-of-the-art methods. The results proved that our Ts-DRL has greatly promoted the development and deployment of contrast-free liver tumor technology.
Link to paper
DOI: https://link.springer.com/chapter/10.1007/978-3-031-16443-9_26
SharedIt: https://rdcu.be/cVRyE
Link to the code repository
N/A
Link to the dataset(s)
N/A
Reviews
Review #1
- Please describe the contribution of the paper
A ternary knowledge transferred teacher-student DRL (Ts-DRL) for liver tumor detection that is contrast-free (without the use of chemical injection) which is safe, speedy, and inexpensive technology.
- 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.
- Paper has good and strong evaluation dataset - 352 patients with data from two MR scanners which helps to prove that the technique proposed has strong test value.
- The formulation of the technique is simple but good enough to be understood in layman term ; Ts-DRL provides knowledge set that is ⟨A,R,F⟩ - action, reward, and feature from the DRL’s agents of teacher network to guide the student network learning.This ternary knowledges also inform the student network what to do,and innovatively embeds F to teach the student the reason and purpose behind the A,R.
- The novelty of this method is the why to do of the ternary knowledge framework (A,R,F) .Thus, it improves the effectiveness of the teacher-student DRL framework and the accuracy of detection. It also possesses a novel progressive hierarchy transferring strategy (P-strategy) to strengthen the transfer of ternary knowledge.
- 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.
- In testing phase, well trained student network drop the ternary knowledge from the teacher network. Why the student get rid of the knowledge from the teacher network? What method is used? Without the teacher network, the student will not have the previous knowledge plus with the new knowledge. Is this process necessary?
- Experimental ablation is used in research on animals. Is this experiment necessary for the human liver datasets?
- 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
Most reproducibility checklist are fulfilled.
- 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
- Paper has good organization and easy to follow experiments used for getting the right results.
- Results were explained in a manner where non-scientist person can understand the process of the method and result analysis used to prove the technique easy to follow.
- 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?
- Author was able to justify the method used with good explanation in methodology and structured analysis.
- Ternary knowledge add another feature F where the student get to know why the action need to be done.
- Number of papers in your stack
2
- 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 #2
- Please describe the contribution of the paper
This paper presents contrast-free liver tumor detection using ternary knowledge transferred teacher-student deep reinforcement learning. The test results are also provided.
- 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 topci of this paper is interesting. The methodology used in this work is well described.
- 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 jobs should be done to better verify the work.
- 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 reproducibility of the paper is OK.
- 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
Here, there are some review comments of this reviewer:
1 The computational cost of the proposed approach isn’t discussed. The approach should be computationally efficient to be used in practical applications. 2 The proposed method might be sensitive to the values of its main controlling parameter. How did you tune the parameters? 3 Have you considered the effect of noises on the performance of the proposed method? Please discuss how this would impact the results and conclusions of this study. 4 The practicality of the approach should be further discussed.
- 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 topci of this paper is interesting. The paper is technialy sound with well presentation.
- Number of papers in your stack
4
- What is the ranking of this paper in your review stack?
2
- 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 #4
- Please describe the contribution of the paper
This paper enhance the previous work WSTS by integrating additive teacher model features into the teacher-student based DRL framework with a novel P-strategy. Experiments show that the proposed method outperforms previous baseline methods.
- 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 method is well motivated to use a P-strategy to stale the training process which is better than other DRL strategy including DQN, DPG, DDPG, etc.
- The extensive experimental results show that the proposed methods can detect tumor better than previous baseline models.
- 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 may not easy to reproduce the results, because some details are missing:
- how to select hyper-parameter in the reward function, grid searching?
- for the proposed P-strategy, since P-strategy only introduces one item (4 items in total) from the student network’s features that is most similar to the teacher at each step, how to decide when shoud perform the 4 exchangement to make the training process stable?
- 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
It may not easy to reproduce the results due to some missing details and the in house dataset.
- 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
- More experiment details can be provided for the reproducibility.
- The resolution of Fig 4 can be improved for visual comparision. The regions of interest can be zoom in.
- 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?
Although some details are missing, the paper study an important problem and the proposed method show improvement than previous baselines. This work will interest more researchers and clinical doctors if the datasets and source codes will be availabel.
- Number of papers in your stack
8
- What is the ranking of this paper in your review stack?
2
- 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.
The paper presents a novel ternary knowledge transferred teacher-student Deep Reinforcement Learning method for contrast-free liver tumor detection. The reviewers were in agreement with the importance of the topic, the novelty of the work presented, as well as the clarity of the presentation. Minor concerns were raised regarding the presentation of the experimental results, such as description of the methods under comparison and the baseline methods, which I believe can be successfully addressed in the final 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).
2
Author Feedback
N/A