Attentive Octave Convolutional Capsule Network for Medical Image Classification

Round 1

Reviewer 1 Report

The authors have improved the quality of the manuscript and addressed my comments. however, there are some minor comments:

1- for the performance metrics equations please define N, TP, TN, FP, FN.

2- please also define macro precision and macro recall.

3- In tables, 9-12, could u please at least mention the performance values u attained in the text.

Author Response

Dear reviewer,

Thanks for taking your precious time to review this manuscript. We have thoroughly revised the manuscript based on your comments. The revised texts are marked in red in the manuscript.  Please refer to the attached file for our responses point by point.

Author Response File: Author Response.docx

Reviewer 2 Report

The author proposed a novel framework, named AOC-Caps, based on the combination of a capsule network (CapsNet) and an attentive octave convolutional AOC module, for medical image classification. 

Overall, the paper is well written and quite easy to follow and understand. The authors already integrated the suggestions of other reviewers improving the quality and understandability of the suggested framework. I only suggest improving the Related Work section by including some well-known medical image classification framework. I suggest discussing other state-of-the-art methods. Some possible references are: [Isola et al. "Image-to-image translation with conditional adversarial networks", 2016], [Ronneberger et al. “U-net: Convolutional networks for biomedical image segmentation”, 2015], [Milletari et al. "V-Net: fully convolutional neural networks for volumetric medical image segmentation", 2016], [Badrinarayanan et al. "SEGNET: a deep convolutional encoder-decoder architecture for image segmentation", 2017], [Pelt and Sethian "A mixed-scale dense convolutional neural network for image analysis", 2018], [Rundo et al. "USE-Net: Incorporating Squeeze-and-Excitation blocks into U-Net for prostate zonal segmentation of multi-institutional MRI datasets", 2019], and [Rundo et al. "CNN-based prostate zonal segmentation on T2-weighted MR images: a cross-dataset study", 2020].

For all the aforementioned reasons, I suggest a minor revision of the manuscript.

Author Response

Dear reviewer,

Thanks for taking your precious time to review this manuscript. We have updated the related work section. Please refer to our revision report as attached for details. 

Thanks.

Author Response File: Author Response.docx

Reviewer 3 Report

My congratulations. The article is very interesting and presents new ideas.  The bibliography is up-to-date and well-chosen. A large number of experiments is an important element of the work.

Author Response

Dear reviewer,

Thanks for taking your precious time to review this manuscript.  We really appreciate your feedback.

Reviewer 4 Report

The paper presents a method for medical image classification based on capsule net and attentive octave convolution to process and combine high and low frequency information simultaneously within the network. There are few comments and concerns (as follows) which needs to be addressed before the paper can be accepted.

1. The statement (“When these parts are disturbed, CNNs would still recognize the disturbed object as a bird, while CapsNet can determine that it is not a bird through the part-whole relationship”) is not convincing. There needs to be an example or reference to prove the validity of the statement
2. “lower frequencies correspond to the global information in the images”: what is meant by global information here? The texture in an image can also mean global information but are definitely not low frequency components.
3. “Similar to the MNIST dataset [20], MedMNIST performs classification tasks on images” – is MedMNIST a method or a dataset? A dataset cannot perform classification task.
4. “…. two of the seven tasks” What are the seven tasks here? Only four are mentioned in the experiment section
5. Why are m⁺ and m^- chosen as 0.9 and 0.1?
6. Is there a way to show how the method performs across staining procedures for histopathological dataset?
7. What are macro-precision and macro-recall?
8. Please mark the highest accuracy/precision and recall in the table in bold for better understanding

Author Response

Dear reviewer,

Thanks for taking your precious time to review this manuscript. We have thoroughly revised the manuscript based on your comments. The revised texts are marked in red in the manuscript. Please refer to our revision report as attached for our responses point by point.

Author Response File: Author Response.docx

Reviewer 5 Report

I wouldn't add anything else!
Good job!

Congratulation!

Author Response

Dear reviewer,

Thanks for taking your precious time to review this manuscript. We really appreciate your feedback.