Next Article in Journal
An Integrated Approach to Determining the Capacity of Ecosystems to Supply Ecosystem Services into Life Cycle Assessment for a Carbon Capture System
Next Article in Special Issue
A High-Accuracy Mathematical Morphology and Multilayer Perceptron-Based Approach for Melanoma Detection
Previous Article in Journal
Learning Collision Situation to Convolutional Neural Network Using Collision Grid Map Based on Probability Scheme
Previous Article in Special Issue
Quantification of Liver Fibrosis—A Comparative Study
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Applied Sciences
Volume 10
Issue 2
10.3390/app10020615
Review Report
applsci-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
Article
Peer-Review Record

Enhancing Multi-tissue and Multi-scale Cell Nuclei Segmentation with Deep Metric Learning

Appl. Sci. 2020, 10(2), 615; https://doi.org/10.3390/app10020615
by Tomas Iesmantas 1,†, Agne Paulauskaite-Taraseviciene 2,*,† and Kristina Sutiene 3,†
Reviewer 1: Anonymous
Reviewer 2: Anonymous
Appl. Sci. 2020, 10(2), 615; https://doi.org/10.3390/app10020615
Submission received: 19 December 2019 / Revised: 6 January 2020 / Accepted: 8 January 2020 / Published: 15 January 2020
(This article belongs to the Special Issue Image Processing Techniques for Biomedical Applications)

Round 1

Reviewer 1 Report

This paper presents a deep learning method for Cell Nuclei Segmentation with a new loss term. The Cell Nuclei Segmentation is a very challenging task, and the method indeed can improve segmentation accuracy.

 

Comment-1:

"Typically, most of CNN architectures use adopted cross-entropy based softmax loss function"

Change " adopted cross-entropy based softmax loss " to "cross-entropy loss" which is enough

 

Comment-2:

Please add the reference papers for " focal loss" and " Tversky loss "

 

Comment-3:

"DW(G(X1), G(X2)) is some distance metric for a pair of input"

Please explain the distance metric. Is it Euclidian distance or something else?

 

Comment-4:

"while G(X1) and G(X2) are two vectors generated as a new representation of paired samples in embedding space."

Please explain G(X1): where does G(X1) come from?   Some layer of the network? I assume it refers to "embeddings of those M pixels from the 12th layer."

 

Comment-5:

Please give more information about training, validation and testing sets, .i.e. how many images per set.

" 5-fold cross-validation"- is it performed on training set ?

 

Comment-6:

About Figure 2, why not reports DICE of all five cross validation sets like Table 2?

 

 

The link does not work

" Supplementary Materials: The code developed during this study is available at [name of repository] https://www.mdpi.com/2076-3417/xx/1/5/s1."

 

 

 

 

 

Author Response

Please, find the response letter attached.

Author Response File: Author Response.pdf

Reviewer 2 Report

Authors present a novel segmentation technique for nuclei segmentation. They have studied the performance of CNN using different metrics. Their overall system achieves an accuracy of 0.13% and 22.31%. However, I would recommend them to study the performance U-Net and add the same in the comparison table. Below is a paper on U-Net segmentation used for X-ray segmentation. Please study the performance of the same and see how the results. U-Net segmentation have proven to be highly effective for Bio medical imaging applications. 

- https://www.researchgate.net/publication/332589403_A_Computationally_Efficient_U-Net_Architecture_for_Lung_Segmentation_in_Chest_Radiographs

 

Author Response

Please, find the response letter attached. 

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

thanks for answering my questions

Back to TopTop