An Investigation of Transfer Learning Approaches to Overcome Limited Labeled Data in Medical Image Analysis
Round 1
Reviewer 1 Report
The paper presents a new method based on transfer learning to overcome limited labeled data. The paper is written and organized well. The methods and results are clearly explained and confirm the novelty of the work.
Author Response
Thank you for the review of our manuscript.
Best regards
Jinyeong Chae
Reviewer 2 Report
1- The motivation is not clear. Why did this work? Is any problem does it address that the previous methods could not?
2- Please mention the contributions in bullets at the end of the introduction section.
3. Why did you use a dataset model?
4. Why did you use the ResNet50 model?. Was there a specific reason?
5. Specify the number of segmentation model parameters in each layer
6- Related work section is poor.
7- include a comparison table in the related work that should highlight the strengths and weaknesses of the proposed method as well as previous methods.
8- some state-of-the-art papers on CNN should be taken into account are:
https://ieeexplore.ieee.org/abstract/document/9107128/
https://link.springer.com/chapter/10.1007/978-3-030-98253-9_20
https://eudl.eu/doi/10.4108/eai.12-4-2021.169183
https://content.iospress.com/articles/journal-of-x-ray-science-and-technology/xst210910
https://ejrnm.springeropen.com/articles/10.1186/s43055-021-00524-y
9- Novelty of the algorithm needs to be incorporated
10- How to deal with overfitting in your model?
11- Besides language also needs improvement.
Besides language also needs improvement.
Author Response
Dear reviewer,
We modified our manuscript and added contents according to the reviewer’s comments.
Also, we wrote our response to the review comment with the arrow symbol as below.
Kindly read our response and manuscript.
Best regards
Jinyeong Chae
Author Response File: 
Author Response.docx
Reviewer 3 Report
In this manuscript, the authors investigated the approaches to overcome the lack of data for representative medical imaging tasks using transfer learning technologies. The tasks were divided into image classification, object detection, and segmentation, commonly needed functions in medical image analyses. They proposed transfer learning approaches suitable for each task that can be applied when there is little medical image data available. The manuscript is written well and easy to follow. I have a few comments before accepting for possible publication.
What is the apparent motivation of this study? It should be justified precisely.
What is the significant contribution of this study?
How is the proposed method differ from the SOTA models?
Improve English writing and remove typos.
Small changes require.
Author Response
Dear reviewer,
We modified our manuscript and added contents according to the reviewer’s comments.
Also, we wrote our response to the review comment with the arrow symbol as below.
Kindly read our response and manuscript.
Best regards
Jinyeong Chae
Author Response File: Author Response.docx
Reviewer 4 Report
Some of the references should be updated with the most recent ones. For example, the first two sentences in Section 1 (Introduction) are about the importance of deep learning and CNN and the cited reference #1 is for 1999. The authors are advised to use the most recent references.
The issues highlighted in Figure 1 are similar and can be merged and reworded. For example; "Limited Data Collection and lack of dataset training", "High cost of labeling and Doctor's diagnosis requirement for labeling"
Under Section 2.3, it is stated "The purpose of image segmentation is to classify all pixels into a specific class." This statement is not valid. The authors should either cite a valid reference to support the correctness of it or should correct it. Any other parts referring to the same meaning must be revised if it's needed.
Transitions from Equation 4 to Equation 5 need to be stated. Moreover, the justification for choosing the Sigmoid function for segmentation output y1 and the Softmax function for classification output y2 should be stated.
Section 5 (Discussion) is written in the form of a summary of the research, while it should discuss the experiments' results and the outputs.
As the existing solutions up to the year 2022 are considered in this research, the authors are advised to explore the latest similar published works up to the year 2023 to ensure this research is up to date.
Proofreading is required. Some sentences in the manuscript are vague and need to be revised. For example, in the abstract, it is written: "...deep learning methods used in existing studies have had a disadvantage in that the number of training samples is insufficient and the labelling cost is high". or "Training approaches considering the common characteristics of medical images are needed." or "The 13 approaches were compared with state-of-the-art results", etc.
Author Response
Dear reviewer,
We modified our manuscript and added contents according to the reviewer’s comments.
Also, we wrote our response to the review comment with the arrow symbol as below.
Kindly read our response and manuscript.
Best regards
Jinyeong Chae
Author Response File: Author Response.docx
Reviewer 5 Report
1. Highlight the aim and novelty of the article
2. Manage the organization of the manuscript
3. Improve the quality of the figures and tables
4. Include the recent references
Need to improve
Author Response
Dear reviewer,
We modified our manuscript and added contents according to the reviewer’s comments.
Also, we wrote our response to the review comment with the arrow symbol as below.
Kindly read our response and manuscript.
Best regards
Jinyeong Chae
Author Response File: Author Response.docx
