Next Article in Journal
An Efficient, Anonymous and Robust Authentication Scheme for Smart Home Environments
Next Article in Special Issue
Feasibility of Social-Network-Based eHealth Intervention on the Improvement of Healthy Habits among Children
Previous Article in Journal
An Identity Authentication Method of a MIoT Device Based on Radio Frequency (RF) Fingerprint Technology
Peer-Review Record

Detecting Respiratory Pathologies Using Convolutional Neural Networks and Variational Autoencoders for Unbalancing Data

Sensors 2020, 20(4), 1214;
Reviewer 1: Anonymous
Reviewer 2: Anonymous
Reviewer 3: Anonymous
Sensors 2020, 20(4), 1214;
Received: 20 January 2020 / Revised: 16 February 2020 / Accepted: 21 February 2020 / Published: 22 February 2020
(This article belongs to the Special Issue Sensor and Systems Evaluation for Telemedicine and eHealth)

Round 1

Reviewer 1 Report

A data augmentation strategy with variational autoencoder is adopted to classify different kinds of pathologies.
I have the following suggestions and concerns:
1. Remove the wordy descriptions about data normalization. Just give what the authors adopt using some brief words because data normalization is very trival.
2. Line 150-151: It is not an intrinsic and relevant descriptions about variational autoencoder. Variational autoencoder is a probabilstic model that learns the data distributions.
Authors should capture the heart of variatioanl autoencoder.
3. The performance index is suggested to use the words that are familiar to readerships, such as recall, precision, and accuracy.
4. Figure 7 occupies a lot of space. It is suggested to use Table instead of the figure.
5. Recently, variatioanl autoencoder is widely used to detect different kinds of signals. But the paper doesn't make a sufficient review. The followings are some examples:
Lee, Seulki, et al. "Process monitoring using variational autoencoder for high-dimensional nonlinear processes." Engineering Applications of Artificial Intelligence 83 (2019): 13-27.
Wang, Kai, et al. "Systematic development of a new variational autoencoder model based on uncertain data for monitoring nonlinear processes." IEEE Access 7 (2019): 22554-22565.
Zemouri, Ryad, et al. "Deep Convolutional Variational Autoencoder as a 2D-Visualization tool for Partial Discharge Source Classification in Hydrogenerators." IEEE Access (2019).

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

This work is interesting to the readers of sensors. Some minor comments:

Line 1: "The aim of this paper is the detection of pathologies in respiratory sounds.", the aim should be written in the past. Instead of "in respiratory sounds", use "through respiratory sounds".

Line 1: ICBHI is not defined.

Line 2: Sample is quite unbalanced?What this means? Abstract needs to be clear. The sample are respiratory sounds, are patients/subjects?

Lines 14-17: Use lowercase to disease names. AUthors should clearly state the use of a respiratory sound recordings dataset. Respiratory dataset is vague.

Lines 18-42: AUthors should present the similarities and differences between diseases in one or two paragraphs, so readers would understand better the importance of the work. One paragraph for each disease is not adequate.

Line 60: A recent work published in sensors also used CNN to detect breathing phases and could be also cited here.

Lines 137- 140: Data normalization was performed with all alternatives presented?It is not clear. In line 192 it seems that only Min_max feature was used.

Discussion of the results and limitations of this work need to be expanded.


Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

The automated detection of respiratory pathologies is a challenge. However, less clear is whether it is also a clinical challenge. For example, it is usually not an issue to detect that somebody has a chronic (respiratory) disorder, as chronic means by definition that the patient has the disease already for some period and that it will never go away. A similar conclusion can be drawn for the classification in six categories, where the detection of COPD, bronchiectasis is clinically nonrelevant. Clinically relevant is the diagnosis of an acure respiratory disease, such as pneumonia, bronchiolitis, bronchitis, or the detection of an exacerbation of COPD and asthma.

Althougth the experimental results show that the results obtained using convolutional neural networks gives superior results, but that is due to the fact that you experimented with architectures and settings of the neural networks until obtaining better results than others. No attempts are made to show that the results obtained generalise beyond the ICBHI database used. Also in the eveluation no use if made of cross-validation or an independent respiratory sound database. Such experiments are needed to justify the conclusion that your method is superior to others. In addition, it is useful to include results of neural networks with simpler architectures for comparison.


Finally, note that your definition of the mean (equation (2)) is actually the variance.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Back to TopTop