COVID-19 Lesion Segmentation Using Lung CT Scan Images: Comparative Study Based on Active Contour Models

Round 1

Reviewer 1 Report

This paper investigates the effectiveness of active contour models for segmenting the area of pneumonia caused by COVID-19, with a focus on the external force based performance. The paper is easy to follow and the topic is of interest for researchers and practitioners in related fields, but should be further polished before being considered for publication. My major concerns include:

1. Although it is true that deep learning usually relies on large training datasets, it would be debatable to say "...there are not enough images to use machine learning methods" in Abstract, since there have been a lot of unsupervised learning techniques.
2. As a review paper, recent lesion segmentation methods using contour models should be mentioned, e.g., "Cerebral Microbleed Detection Via Fourier Descriptor with Dual Domain Distribution Modeling" in IEEE ISBI Workshops 2020.
3. The writing of the paper could be improved and extensive editing of English language is required. Just to suggest a few changes for example: in line 20, "is via viral nucleic acid detection"->"is viral nucleic acid detection"; line 233, "as follow"->"as follows"; line 234, "Powerful for noise"->"robust against noise".

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

In this paper, the authors aim to test different active contour models (ACM) by evaluation on a publicly available CT dataset of COVID patients by comparing segmentation results with other common articial intelligence methods. Although the approach is very exciting and promising, the paper unfortunately lacks structure. The usual sequence of Introduction, Methodology, Results, Discussion was not followed. Instead, Introduction and Methodology are mixed and divided into 4 sections and take up 14/23 pages, which is far too much for an article. Partly the paper seems more like a review than a presentation of own results. In the Introduction there are also some linguistic deficiencies and unfortunate citations - while countless text passages are cited for each methodology, there are large passages without any text reference. For COVID-19 papers, an epidemiological statement - can also be very short - is also essential. The actually new part with Results and Discussion is also mixed and unfortunately comes much too short. Another criticism is that the weaknesses of the ACM methodology are not discussed.
I think that the authors can write a nice paper from the results. However, a thorough revision is needed.

Abstract:

• The abstract seems a bit long.
• Background, Aim, Methods, Results, Conclusion and Outview seems to be mixed up. Authors should follow the common structure

Keywords:

• It is called Chest CT

Introduction:

• The medical information in the first paragraph must be supported by references. Also an epidemological statement, including current infection rates and number of deaths should be included in a COVID paper. Authors can use the data provided by WHO.
• Line 23: Don't write "this" RT-PCR molecular method - it is a common medical technique also used for other viral diseases.
• Line 25: I think the statement that false negative cases of RT-PCR tests are THE potential threat to public wellness is a bit easy too say and does not cove the whole epidemiological truth as governments start to think about using other pandemic indicators than just incidence such as hospitalization. Also note that CT scans have a high false negative rate, too.
• Line 27: CT scans are no alternative to RT-PCR at this state of the pandemic. It might be a valuable supplemental diagnostic tool, e.g. for infection process Evaluation, but one would rarely diagnose covid based on CT alone.
• Line 35: I would not rely on a single paper in terms of the evaluation time. 21.5 minutes seems to a be a quite long time span for just saying that COVID-typical alterations are present or not.
• Line 49: I would say, might be very crucial. Depends on the pathology. COVID-typical alterations cover usually greater areas of the lung.
• Very detailed, but good understandable overview about ACM methods.
• Line 64: ...that's why - I would preferably start with a new sentence
• Line 70: Citation missing - data that shows that ACM are usually slower than parametric
• Line 70: Furthermore,
• Most of the last two paragraphs should be placed in a Methodology section which is missing. Also the last sentence should state a clear main goal and needs a shortening. The three bullet points belong in the text.
• Some language issues in the Introduction section: e.g. Line 96: one hundred COVID-19 CT images
• Line 26: It is very important to figure out how segmentation took place and how the ground truth is defined (Methodology).

Section 2:

• Section 2 is way too long and not necessary. Literature review belongs in the Introduction.
• The first paragraph is missing citations.
• Some language issues.
• 135: repitation. Already stated in Introduction.
• Page 5 is poorly structered. In most parts the paper seems like a Review. Results of other papers are described too detailed.

Section 3:

• Very detailed mathematical explanations.
• Should belong in a Methodology section.
• Figures 3/4 are very descriptive and neat

Section 4:

• Should be included in a Methodology section.
• Very detailed, Figures likewise neat

Section 5:

• This is claimed to be the results section but there is a mix up with methodologic and introductional sequences.
• 5.1. should be part of the Methodology section. Line 307: First sentences are already stated before.
• Results of the figures should be briefly stated, e.g. AUC in Figure 7.
• Figure 6: Please state what is the ground truth
• Line 346: Typo: Table 3
• 5.4. It is a mix of results and discussion. The key results are shown in Table 3 and Figure 9. This part must be put focus on and results need to be extracted from the interpretation which should be part of a Discussion section.
• Missing limitation section!

Conclusion:

• Line 394: The methodology section was not briefly compared to that discussion of advantages, disadvantages
• Line 399: why did the authors not test deep learning and clustering?

Authors should include a section stating the individual contributions.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The manuscript has been sufficiently improved to warrant publication in Applied Sciences.

Reviewer 2 Report

The manuscript has been thoroughly revised and is now in a publishable form.