Segmentation of Echocardiography Based on Deep Learning Model
, Hairui Wang 4, Jing Wu 1, Chunqiang Hu 2,3, Nan Li 1, Xiaomei Wu 1,4,5,6,7,* and Cuizhen Pan 2,3,*
Round 1
Reviewer 1 Report
This paper investigates the classification of mitral regurgitation by a deep learning network VDS-UNET. The author may need to add more machine learning discussion due to this paper is a machine learning-related paper, then this paper can be published in the electrics.
The author may need to add more figures and discussion about the training loss history and regression map.
What's the comparison between the real data and the predicted results?
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Based on a careful analysis, I can formulate the following remarks:
- In my opinion, the main question addressed by the authors consists in putting in evidence the importance of a new type of an efficient-, easier-, as well as a quick evaluation strategy of the video echographies in order to evaluate the different severity of the mitral regurgitation disease.
2. The topic represents in my opinion a relevant approach of the proposed theme in the field of heart diseases. Even there are some (a relative few number of) similar researches, their proposed approach/methodology come to fill a specific/particular gap in the mentioned field.
-
In comparison with other published material, the authors' contribution adds to the subject area an new deep-learning network, that can assure a simultaneous and accurate segmentation achieving in multi-section echocardiography and so assuring a further efficient and precise quantitative measurement of clinical parameters with respect to mitral regurgitation.
The obtained results were better than the other reported ones from the literature.
-
About the specific improvements which the authors consider regarding the methodology as well as the further controls which should be considered, I can mention the following:
It is well-known fact that the mitral regurgitation (MR) represents one of the most common and also dangerous heart diseases. Its clinical treatment differs in dependence of the type of MR and its severity. Consequently, became very important to evaluate correctly the achieved echocardiograms videos; the individual/manual evaluation/analysis of the videos requires high-qualified cardiologists and there are very time-consumptions activities.
In order to assure an unified standard, intelligent and efficient auxiliary decision-making system in clinical praxis, i.e. to improve the accuracy and efficiency of the MR diagnosis, several studies were performed.
The authors designed a high efficient deep-learning network in order to perform a classification of mitral regurgitation (MR) from echocardiography.
In this sense they started their researches from a large expert-labelled dataset of echocardiograph videos (without and with different severity of diseases/dysfunctions), than applied the suitable convolution network in order to replace the contraction path in the original UNet work to extract image features and added depth supervision to the expansion path to achieve the desired segmentation.
The authors proposed and tested clinically a combined UNet and VG16 networks in order to achieve simultaneous segmentation of the significant zones, obtaining finally an improvement of the actual deep-learning methods.
- In my opinion, the presented conclusions are suitable related to their research results and prove that they reached the proposed goal.
- The references in my opinion are appropriate.
- With respect to the presented figures and tables I can mention:
Of course, even their contribution has a page-number limitation; with the Editor’s acceptance they can include some supplementary figures, where different specific cases, i.e. severity of diseases/dysfunctions can better illustrate the differences in comparison with the healthy heart’s images.
In similar manner, the tables can be also enlarged in order to put more detailed in evidence the proposed methodology’s efficiency.
Even they will not add any supplementary figures or tables, the presented form of this contribution in my opinion can be published.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
