Authors

Jhih-Ciang Wu, Ding-Jie Chen, Chiou-Shann Fuh

Abstract

Machine learning based Computer-Aided Diagnosis (CAD) aims to assist clinicians in the pathological diagnosis process. While dealing with video pathological diagnosis such as colonoscopy polyp detection, the recent SOTA method employs Weakly-supervised Video Anomaly Detection (WVAD) in the Multiple Instance Learning (MIL) scenarios to concern the temporal correlation within data and to formulate the concept of the interest disease simultaneously. Such a MIL-based WVAD method leverages video-level annotations to detect frame-level diseases and shows promising results. This paper casts the video pathological diagnosis as a MIL-based WVAD task and introduces Contrastive Feature Decoupling (CFD) network to decouple normal and abnormal feature ingredients per snippet. With such decoupled features, we are able to highlight the abnormal feature ingredients for accurately reasoning the disease score per snippet. The core components within our CFD model are the memory bank and contrastive loss. The former is used to learn atoms for representing normal features, and the latter is used to encourage our model to gain robust disease detection. We demonstrate that our CFD network is achieving new SOTA performance on the existing Polyp dataset and the introduced PANDA-MIL dataset. Our dataset are available at https://github.com/Jhih-Ciang/PANDA-MIL.

Link to paper

DOI: https://doi.org/10.1007/978-3-031-43904-9_25

SharedIt: https://rdcu.be/dnwG3

Link to the code repository

https://github.com/Jhih-Ciang/PANDA-MIL

Link to the dataset(s)

https://github.com/Jhih-Ciang/PANDA-MIL

Reviews

Review #3

Please describe the contribution of the paper

This paper applies the concepts of Kinetic dataset (Carreira and Zisserman) to the problem of detection on the polyp and prostate biopsies databases.
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 paper compares algorithms and shows that the proposed algorithm outperforms other algorithms on the same datasets. It combines ideas from other areas like vide analysis for medical datasets. Evaluation includes ablation analysis and comparisons with other published 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.

The paper could be better written, there are typos (liner for linear), some maths are not well defined (like the parameter C) and the results most important, those related to the classifications are not very clear, i.e. what is the frame index of Fig 2?
Please rate the clarity and organization of this paper

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

Datasets are open from other studies Code is supposed to be in Github and not available due to anonymity
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

I think that the paper would be better framed only as a medical imaging problem and not trying to generalise to video/snippet/WSI, which is probably the source of some unclarities.

Fig 2 is the most important one, and is not that clear, it would be better to present a WSI, zooming into and show the areas that are healthy or diseased, or how the algorithm classifed the whole WSI.

There is one point that I am not convinced, how can something have a 99.51 AUC and yet only 88.13% in accuracy? To get to 99% the FP and FN must be really few as compared with TP and TN and then it is hard to see how you can get 88 accuracy.
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?

Paper is Ok, could be better written and I have a worry about the calculations. This could be OK as a poster.
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 #1

Please describe the contribution of the paper

This paper proposed a contrastive feature decoupling network, introduced a new dataset for prostate cancer detection, and demonstrated SOTA performance on two biomedical imaging datasets.
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 main strengths of the paper include an original way to use data, and the demonstration of a novel application of the proposed contrastive feature decoupling network in pathological image classification.
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.

Although there are some novel aspects of the proposed network, the paper presents an incremental advance over previous work.
Please rate the clarity and organization of this paper

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

In the abstract, the authors indicate both code and dataset will be made available through Github.
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

This paper presented important research on topic of broad significance. The proposed method has some novel aspects. The experimental design and evaluation of the data are satisfactory, and conclusions are justified. The paper is well written in clear English and is easy to follow. However, there are some grammar errors and typos which need to be corrected. For example, “We introduce the new biomedical imaging dataset of prostate cancer detection, i.e., PANDA-MIL” should be “We introduce the new biomedical imaging dataset for prostate cancer detection, i.e., PANDA-MIL”. While the experimental results show the outstanding performance of the proposed CFD network, the paper did not provide enough analysis on why the CFD network works better than other methods listed in Table 2.
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?

This paper presented a new imaging dataset of prostate cancer detection, proposed a contrastive feature decoupling network, conducted extensive experiments and showed SOTA performance of the proposed network.
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.
This paper focuses on Colonoscopy polyp detection combining Multiple Instance Learning and Wealy supervised video anomaly detection.

The strengths of the paper include:
- use of contrastive feature decoupling (used in consecutive frames).
- show the utility of the proposed approach for two very different tasks (polyp detection and prostate cancer in biopsies).
- Great visualization in figure 2.
The paper should undergo careful proofreading, (e.g., expertise-required in the introduction), including not using the term videos for panda dataset.

Author Feedback

N/A

back to top

Contrastive Feature Decoupling for Weakly-supervised Disease Detection