Machine Learning Driven Contouring of High-Frequency Four-Dimensional Cardiac Ultrasound Data
Round 1
Reviewer 1 Report
The manuscript, entitled "Machine Learning Driven Contouring of High-Frequency Four-Dimensional Cardiac Ultrasound Data," demonstrates the 4DUS boundary detection using machine learning. The method and obtained results are well described, then this manuscript can be suggested for publication after minor revise as below.
In section 2.1
Please add the frame rate and time duration of the obtained 4DUS images.
In Fig.1,
Please display the coordinate system.
At line 198,
R2 should be R2.
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
This submission presents an automatic boundary detection method from 4D ultrasound (4DUS) cardiac imaging data on mouse models. In particular, the following three models were proposed: (1) short-axis images analyzed individually; (2) parallel slices incorporated simultaneously; (3) predictions assisted by a single user-input position.
The proposed approach is interesting and the methodology overall sound. The manuscript is well prepared.
However, some improvements would be beneficial to increase the quality and clarity of the proposed work. In particular, the contributions with respect to the state-of-the-art might be emphasized.
My main concerns are listed in what follows.
1) Abstract needs rewriting.
More clarity would be beneficial in the description of the three Models – the list (Model 1, Model 2, Model 3) is not very readable for Abstract. More importantly, quantitative results have to be better outlined while pointing out the main achievements.
Moreover, the English language would be revised. For instance: 'we have developed and demonstrate here' -> 'we have developed and demonstrated here'
2) Keywords: more specific keywords should be added, such as: '4D ultrasound'.
3) Section 1: the structure of the manuscript should be outlined at the end of the Introduction.
4) The literature background might be extended, possibly in a separate Background (or Related Work) section. This will allow the Authors to better clarify the actual contributions of the proposed work.
5) Regarding the importance of epicardial fat segmentation on Computed Tomography (CT), please consider to introduce and discuss the following recent approaches on classic image processing [Militello, C., Rundo, L., Toia, P., et al. (2019) A semi-automatic approach for epicardial adipose tissue segmentation and quantification on cardiac CT scans. Computers in Biology and Medicine, 114, 103424. DOI: 10.1016/j.compbiomed.2019.103424] and deep learning [Commandeur, F., Goeller, M., Razipour, A., et al. (2019) Fully Automated CT Quantification of Epicardial Adipose Tissue by Deep Learning: A Multicenter Study. Radiology: Artificial Intelligence, 1(6), e190045. DOI: 10.1148/ryai.2019190045].
6) Statistical treatment would be improved. In particular, the use of parametric Student's t-test needs
7) Figure 3 is complicated for interpretation. More comments would be beneficial.
8) Section 5: Conclusions should be extended and future work needs an actual feasibility plan supported by relevant literature references.
9) A few statements on the applicability of the proposed approaches to human models would be interesting.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report
The Authors have carefully addressed the issues raised in the first revision round. The manuscript is now in better shape.
Especially, the Abstract is comprehensive and provides a good view of the proposed work.
The rationale and practical relevance, including the state-of-the-art (with appropriate references), have been clarified.
Some improvements to the manuscript structure and style have been performed.
However, there are additional improvements to be taken into account by the Authors.
1) I would suggest avoiding structured Subsections in Section "1. Introduction".
Possibly, the related work in Section Background might be moved to a separate Section 2.
2) Many thanks for understanding my previous incomplete comment (apologies for that).
Even though the use of the central limit theorem is reasonable, since the sample sizes used to compute the test statistics were greater than 600 (assuming 30 as a well-known threshold for parametric testing), further justification for the use of parametric Student’s t-test could be beneficial.
For instance, did the Authors consider to use the Shapiro–Wilk test for assessing the normality assumption?
3) Figure 3: the additional description of the figure might be moved to the main text for better readability.
4) Section 3.4: The use of Generative Adversarial Networks (GANs) for data augmentation is a nice future application.
I would suggest to add the following relevant references for GAN-based data augmentation in medical imaging:
- Han, C., Rundo, L., Araki, R., Furukawa, Y., Mauri, G., Nakayama, H., Hayashi, H. (2020) Infinite brain MR images: PGGAN-based data augmentation for tumor detection. In Neural Approaches to Dynamics of Signal Exchanges (pp. 291-303). Springer, Singapore. DOI: 10.1007/978-981-13-8950-4_27
- Xu, J., Li, M., Zhu, Z. (2020) Automatic data augmentation for 3D medical image segmentation. In International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI) (pp. 378-387). Springer, Cham. DOI: 10.1007/978-3-030-59710-8_37
Please consider to introduce and discuss these recent articles for completeness.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
