Breathe out the Secret of the Lung: Video Classification of Exhaled Flows from Normal and Asthmatic Lung Models Using CNN-Long Short-Term Memory Networks
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThis study is a nice and useful example of an experimental (in vitro) approach with deep learning. 3D-printed lung model offered a unique perspective to study changes with controlled parameters.
In the field of computer vision, an occlusion sensitivity map is a technique used to understand which parts of an input image are most important for a neural network's predictions. By leveraging CNN's feature extraction with LSTM’s sequence learning, this study evaluated the feasibility of using exhaled flows to detect constrictive lung diseases and stage the disease severity. Interesting differences were observed regarding the performances of AlexNet-LSTM vs. GoogLeNet-LSTM, video-based vs. image-based classification, and the proposed flow-based diagnostic method vs. previous techniques, as discussed below in detail.
Video classification outperformed image classification at both levels, and the advantages of sequence learning became more obvious when tested on more adverse conditions. This explained the improved accuracies of CNN-LSTM over CNN. It was observed that video inputs result in higher accuracy than image inputs, and the AlexNet-LSTM net work was robust to flow deviations and slightly outperformed the GoogleNet-LSTM net work.
The only comment is do the authors really need that many references?
Author Response
The only comment is do the authors really need that many references?
Response: Following the Reviewer’s suggestion, we evaluated the relevance of each reference and as a result, removed five papers from the reference list.
Reviewer 2 Report
Comments and Suggestions for AuthorsThis research focused on the Video Classification of Ex- 2 haled Flows from Normal and Asthmatic Lung Models Using 3 CNN-LSTM Networks. The following points have to be incorporate in revised manuscript:
1. Related work or literature review section is weak. Author has to include more related study.
2. Author has to provide the objective formulation method.
3. There are minor language correction required.
4. On which basis, author has selected GoogleNet and Alexnet models. Author has to mention the reason behind the selection of used models.
5. Author has to also develop or discuss a mathematical model for the proposed model.
6. Author has to add more parameters for the performance evalution. of the proposed model.
Comments on the Quality of English LanguageMinor language editing required.
Author Response
- Related work or literature review section is weak. Author has to include more related study.
Response: Compared to extensive applications of image-based classification, the application of video classification for disease diagnosis is still at its early stage. This is particularly the case in lung diagnosis using exhaled flows. In this study, we have included 22 references [1-22] for exhalation-based lung diagnosis and 24 references [23-46] for video classifications.
- Author has to provide the objective formulation method.
Response: We thank the Reviewer for the suggestion. Considering that the manuscript is already 20 pages, and the formulation method has been presented in detail in our previous studies, a reference was provided for interested readers. One new sentence was added on page 5, line 199-200: “More details of the methods to calculate the performance matrices can be found in [50].”
- There are minor language correction required.
Response: Following the Reviewer’s suggestion, we have proofread the manuscript carefully and made corrections wherever appropriate.
- On which basis, author has selected GoogleNet and Alexnet models. Author has to mention the reason behind the selection of used models.
Response: The reason behind the selection of GoogleNet and AlexNet was highlighted on page 4, lines 159-163.
- Author has to also develop or discuss a mathematical model for the proposed model.
Response: A reference [48] was given that described in detail a mathematical model for the proposed in vitro model (lines 109-110). The other reference [50] was provided that provided the equations for the calculation of performance matrices (lines 199-200).
- Author has to add more parameters for the performance evaluation. of the proposed model.
Response: In this study, the parameters for the performance evaluation include precision, sensitivity, specificity, F1 score, ROC (receiver operating characteristic) curve, and AUC (area under the curve). In the case of three-class classification (D0, D1, D2), these categorical metrics were adapted from their binary versions, employing the One-vs-Rest (OvR) method, e.g., D0 vs. (D1 and D2), D1 vs. (D0 and D2), and D2 vs. (D0 and D1). The details of the performance matrices were highlighted on page 5, lines 193-199.
Comments on the Quality of English Language. Minor language editing is required.
Response: Following the Reviewer’s suggestion, we have proofread the manuscript carefully and made corrections wherever appropriate.
Author Response File: 
Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsDear authors,
the manuscript is very interesting however, some improvements are required to improve the manuscript quality.
1. At the end of the introduction section add a short description (outline) of the manuscript.
2. Add grids in figures on line 273 (figure 4), 298 (figure 5)
3. Rewrite the conclusion section.
3a. First paragraph a short description of what was done in this paper,
3b. Answers to the hypotheses given in the Introduction section based on discussion given in the previous section.
3c. Pros and cons of your research methodology.
3d. Final paragraph: future direction to improve cons of your research methodology.
4. I did not find in the manuscript train/test methodology. How have you trained your models? Please describe that. Have you used cross-validation which one? 5-fold, 10-fold ... have you investigated the influence of cross-validation on the accuracy of trained models? If not please include it in future research.
Author Response
The manuscript is very interesting however, some improvements are required to improve the manuscript quality.
- At the end of the introduction section add a short description (outline) of the manuscript.
Response: A outline of the manuscript was provided at the end of the introduction.
(Lines 99-103): “The remaining text is organized as follows. In vitro models, experimental setup, CNN-LSTM networks, and study design will be described in section 2. The results of high-speed recording, PIVlab analyses, classification performance, and sequential and spatial features will be presented in section 3. The insights gleaned from this study will be discussed in section 4, with a concise conclusion in section 5.”
- Add grids in figures on line 273 (figure 4), 298 (figure 5)
Response: Following the Reviewer’s suggestion, grids were added in Figure 4 (line 283) and Figure 5 (line 315).
- Rewrite the conclusion section.
3a. First paragraph a short description of what was done in this paper,
3b. Answers to the hypotheses given in the Introduction section based on discussion given in the previous section.
3c. Pros and cons of your research methodology.
3d. Final paragraph: future direction to improve cons of your research methodology.
Response: We thank the reviewer for suggesting a structured conclusion. Following the suggestion, the conclusion has been rewritten as follows.
(Lines 635-645): “This study evaluated the feasibility of using video classification of exhaled flows for lung diagnosis by leveraging LSTM’s sequential learning capacity and CNN’s automatic feature extraction. The results showed that video inputs were observed to give higher diagnostic accuracies than image inputs. The AlexNet-LSTM network slightly outperformed the GoogleNet-LSTM network and was robust to reasonable breathing deviations during data acquisition. The union of deep learning and physiology-based in vitro modeling is promising to disclose anomaly-sensitive features amenable to non-invasive lung diagnosis. It was noted that this study used in vitro lung models with constant exhalation, which had the advantage of controlled asthmatic constrictions. Future studies were needed to consider the more physiologically realistic scenarios such as tidal breathing, compliant airways, and varying disease sites.”
- I did not find in the manuscript train/test methodology. How have you trained your models? Please describe that. Have you used cross-validation which one? 5-fold, 10-fold ... have you investigated the influence of cross-validation on the accuracy of trained models? If not please include it in future research.
Response: More details were provided on model training (Methods, lines 212-215): “A ten-fold cross-validation approach was used to train the model, with the dataset randomly divided into ten equal-sized subsets. Ten runs were conducted and in run, nine subsets were used for training and the reaming dataset for testing.”
Author Response File: Author Response.pdf
