Recent Advances in Machine Learning Applied to Ultrasound Imaging
Round 1
Reviewer 1 Report
Summary
The review study focuses on ML methods employed in Ultrasound Imaging mainly focusing on medical diagnostics and non-destructive evaluations. Major part of the review focuses on various ML methods employed in detection, segmentation and classification of human organs, Breast, Arteries, Heart, Liver, Fetus, Lungs and other organs. The minor part of the review focuses on ML methods used in US for non-destructive evaluation for identifying cracks, structural problems in turbine blades, aircraft panels, ceramics etc. The authors also discussed limitations faced by various authors in the literature and highlighted potential areas of future work in conclusions. Overall, this excellent review of ML methods in US imaging sheds light on growing application of ML. The authors efforts in providing a clear and crisp review is greatly appreciated. The manuscript is well-organized and all the methods are described. This manuscript can be accepted after providing some standard references for the ML methods in the methodology section and after minor grammatical corrections (below).
Decision: Minor revision
Suggestion: A table or bar graph indicating different ML algorithms on x-axis and how frequently were they used across all organs or by individual organ would be interesting to see.
Minor Comments
-
Lines 39 – 46: Instead of paragraph, authors can specify as sub-section or section
-
Line 84: “The prediction of the category to which belong...”. Seems like there is a word missing. This sentence need to be revised
-
Line 86: Non-linear SVM results are useful. This sentence can be revised to “Non-linear SVM results are useful when data are not separable linearly” or something similar.
-
Line 87: Rephrase into: “This approach involves implementation of kernel trick [31], a non-linear function which replaces the scalar product....”
-
Line 89: Since the authors are speaking about non-linear SVM, the linear kernel can be removed.
-
Line 98: Please include more information on Random Forests. For example, RF is a bagging method wherein multiple weak learners are trained on bootstrap replicas of the training dataset. More information can be found in Breiman, 2001.
Breiman, L., 2001. Random Forests, Machine Learning. Kluwer Academic Publishers, Manufactured in The Netherlands.
-
Lines 73, 91, 98,103, 112, 120 -> Please provide references for these ML algorithms.
For example for Decision trees, the authors can refer to Breiman 2001, 1998, James 2013
James G., Witten, D., Hastie, T., Tibshirani, R., 2013. Tree-Based Methods. Springer, New York, NY, pp. 303–335. https://doi.org/10.1007/978-1-4614-7138-7_8
Breiman, L., 1998. Classification and regression trees. Chapman & Hall/CRC.
-
Line 108 -> Provide references for the KNN and distance measures.
-
Line 112 -> Rephrase the sentence for better readability. Also provide how LDA is different from SVM.
-
Line 118-> The outputs are labeled.
-
Line 120-> Provide reference for PCA.
-
Lines 164 – 166 -> These are repeated as in lines 17-20. Keep only one of them.
-
Lines 168 – 173 -> Repeated paragraph, similar to the information present in lines 21 – 29. Keep only one of them.
-
Line 175-> Provide reference for pulse-echo method.
-
Line 198-> Provide suitable references for the doppler methods.
-
Line 208 -> cardiovascular analysis is IntraVascular Ultra-sound
-
Line 211-> Provide reference for elastography,
-
Line 319-> AutoML Vision is revealed to be greatly accurate...
-
Line 417-> “made up challenging” can be rephrased.
Comments for author File: Comments.pdf
Author Response
Please fee the attachement
Author Response File: Author Response.pdf
Reviewer 2 Report
The paper is a mere juxtaposition of work done in one area, but to my mind there is no scientific contribution whatsoever.
To begin with, the paper starts with an introduction to machine learning methods that a minimally specialised reader already knows, and which can be found in the most basic machine learning manuals. It seems like a way to fill pages.
On the other hand, the mere enumeration of 200 papers makes the paper unreadable.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 3 Report
In attachment
Comments for author File: Comments.pdf
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 4 Report
This paper provides a review of machine learning applied to ultrasound imaging. The English written is good, while the content of the manuscript seems having limited guidance for researchers in the field of ultrasound imaging processing. Additionally, some other issues should be clarified as follows:
Question 1: The motivation of the paper is unclear. Although it is a review paper that concludes many existing machine learning methods for ultrasound imaging processing, it seems like a repeatable description of relevant references without any personal opinion.
Question 2: Some figures are blurry and unclear. As a result, the figures are suggested to redraw in the revised version.
Question 3: The future work presented in the submitted manuscript has limited guidance for researchers, especially in terms of image algorithms. Almost all of the people know that the combination of ultrasound imaging and deep learning should be explored.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
I insist that the paper does not have the quality to be published in an indexed journal. The authors shy away from improving the paper and have limited themselves to beating around the bush.
Author Response
We just acknowledge the Reviewer’s point of view. We disagree with him and we did not appreciate his way to act as a Reviewer, which seems quite different from commonly recognized practice.
We revised the paper according to the editor’s recommendations. All parts of the paper modified from the previous versions were highlighted in yellow in a specific uploaded file.
Reviewer 4 Report
The necessary revision note is suggested to provide.
Author Response
We revised the paper according to the editor’s recommendations. All parts of the paper modified from the previous versions were highlighted in yellow in a specific uploaded file.