Cascaded Regression-Based Segmentation of Cardiac CT under Probabilistic Correspondences

Round 1

Reviewer 1 Report

In this manuscript, the authors developed a cascaded regression framework for 3D cardiac CT segmentation. By comparing with two other methods, the authors proved the accuracy, effectivity and specificity of this new method.

This is a very nice presentation of this method, and it is well designed. I was wondering if there is any specific requirements of the CT images to use this method. Other than this I don't have further questions. 

Author Response

Thank you for your revision.

1. I was wondering if there is any specific requirements of the CT images to use this method.
=> There is no requirement in the CT images in size and spacing,
because appearance sampling is performed based on physical locations.
By the way, CT protocol can degrade the segmentation if test's protocol is different from training's protocol.

In line 255~256, the following explanation is added to solve your question.
"The input image does not have any requirements in size and spacing, because appearance sampling is
performed based on physical locations."

Reviewer 2 Report

The paper proposes a method for segmentation of heart chambers based on a cascade of regressors and Active Appearance Models.

Although both technique are known for a while, and thus the paper does not contribute on the algorithmic side, as solution (i.e. application to this problem) it is fine.

As criticism:

• the paper compares itself with only one recent paper, and with one from 2002.
• the paper does not provide sufficient parameter ablation
• the English is not smooth on many occasions, although the idea can be extrapolated from the text.

Author Response

Thank you for your comments.

1. the paper compares itself with only one recent paper, and with one from 2002.
=> I added one additional reference paper from 2006 in line 371~375 to strengthen comparison explanation with statistical model-based methods.

<Added Text is shown in following sentences>
"T{\"o}lli et al.'s method performed the segmentation of four-chamber model
by using artificially enlarged training set and Active Shape Model-based segmentation
in 25 subjects. This method produced 1.77 $\pm$ 0.36mm in LV, and 2.44 $\pm$ 0.85mm in LA,
although the enlargement method of training set is used. Our method shows a better result in LA, compared with this method."

Many segmentation researches derived from original active appearance model were performed until 2008 in our view.
From 2008, new methods of Ecabert et al. and Zheng et al's gave high accuracy in segmentation. This explanation is contained in introduction section.
Additionally, the one recent paper of Zhuang et al. is segmentation comparison challenge which contains comparison of 12 algorithms.

2. the paper does not provide sufficient parameter ablation
=> In line 316, initialization part (line 314~318 of previous submission version) in perturbation distribution of shape eigenvector was removed,
since this initialization is composed of routine parameters in SDM segmentation.

<Removed Text is shown in following sentences.>
"and the initial perturbation distribution of the shape eigenvector part ${{\bf{p}}_{k,j}}(t)$ was created from Monte Carlo sampling
of the shape model bounded by $\pm 2\sqrt {{\lambda _i}}$. Additionally, the initial perturbation distribution of
five elements of the similarity-transform part was bounded by fixed values of 400. In the sixth element eigenvector
of z scale, this bound was selected as 600. These two values were selected experimentally."

The remaining parameters have their reasons in manuscript, in our thought.
Since proposed method contains several sub-components, the number of parameters is large.
Sub-components are derived from reference papers. If original papers are understood, the meaning of parameters also can be
caught without difficulty. Moreover, we made a notation section in the first part of paper to help the understanding of parameters.

3. the English is not smooth on many occasions, although the idea can be extrapolated from the text.
=> This paper contains complex mathematical systems. We tried many efforts in removing difficult understanding
when the first submission is performed. If you tell us specific parts, we will consider correction of English.
Since People in South Korea use English as the second language, quality cannot be satisfactory in the perspective of native speakers.
Although we already performed English correction with a native speaker, intrinsic distance can exist from native speakers.
If you request English correction again, we will perform English correction with a native speaker again.

Round 2

Reviewer 2 Report

Although the English is not smooth enough, the idea and results can be understood, thus I recommend publication.