Next Article in Journal
Statistical Validation of Synthetic Data for Lung Cancer Patients Generated by Using Generative Adversarial Networks
Next Article in Special Issue
Deep Learning for Predicting Congestive Heart Failure
Previous Article in Journal
The Influence of a High-Voltage Discharge in a Helicoidal Twisted-Pair Structure on Enzyme Adsorption
Previous Article in Special Issue
Nonlinear Dynamic System Identification in the Spectral Domain Using Particle-Bernstein Polynomials
 
 
Article
Peer-Review Record

End-to-End Decoupled Training: A Robust Deep Learning Method for Long-Tailed Classification of Dermoscopic Images for Skin Lesion Classification

Electronics 2022, 11(20), 3275; https://doi.org/10.3390/electronics11203275
by Arthur Cartel Foahom Gouabou *, Rabah Iguernaissi, Jean-Luc Damoiseaux, Abdellatif Moudafi and Djamal Merad *
Reviewer 1:
Reviewer 2:
Electronics 2022, 11(20), 3275; https://doi.org/10.3390/electronics11203275
Submission received: 23 September 2022 / Revised: 6 October 2022 / Accepted: 8 October 2022 / Published: 12 October 2022
(This article belongs to the Special Issue Machine Learning in Electronic and Biomedical Engineering, Volume II)

Round 1

Reviewer 1 Report

The article is devoted to improving methods for detecting skin lesions in a dermoscopic image by solving the problem of class imbalance in the development of CAD for detecting skin lesions.

The article provides a significant overview of the subject area, in particular, an overview of computer-aided detection systems using neural networks. The authors proposed an improvement in methods for detecting skin lesions using two-stage neural network training based on two new loss functions Lf and Lc.

The proposed method was compared with 7 different methods and the advantage of the proposed method in 3 out of 4 characteristics was shown.

 

Some minor issues:

1. First of all fill authors names and affiliations

2. Figure 2. The text on legend might be larger.

3. In the conclusions, it would be good to add numerical estimates of the results obtained.

Although the study was performed at a high level. With all mentioned, I believe the work may be published.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

1. the caption of Figure 1 seemed incompleted.

2. I was wondering why the authors used 'T0' to name the threshold. First, the subscript, 0, made no sense in this research.  Did it mean there existed T1 or T2? Secondly, the meaning about T0 was closer to be a criteria than being a threshold. In line 365-367 saying, the model would 'switch' the loss function when the accuracy was not improved in ten steps. It is suggested that the author may give a proper symbol for the situation.

3. Due to there existed two differenet loss functions in the training and validate stage. It is suggested that the author may plot the convergence figures, the x-axis should be epoch in training stage. It could give more clear concept about the strategy, more clear than descriptions in line 291-298.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Back to TopTop