Deep Learning of Retinal Imaging: A Useful Tool for Coronary Artery Calcium Score Prediction in Diabetic Patients
, Olga Simó-Servat 2, Alejandra Planas 2,3, Cristina Hernández 2,3, Rafael Simó 2,3 and David Masip 1Round 1
Reviewer 1 Report
In this paper, the author provide a significant evidence that the deep learning can be used for evaluating cardiovascular risk by using CAC as unique biomaker. The paper is well written and presented. However some point the should be taken into account:
1. section 2 is very abstracted. More details about preprosessing, using DNN should , and post processing should be given
2. What about augmentation, it may be a good technique that should be used to improve classification results
3. Sample of training database (images) that used in transfer learning should be shown in a figure.
4. The authors ignored the processing time either in training or in inference, although that they use nvidia tool for GPU as mentioned in the acknowledgment
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The authors proposed the deep learning approach for predicting CAC scores from retinal fundus imaging of diabetic patients.
Here are the comments and suggestions to improve the paper.
- Dataset description is required.
- Abbreviations are missing such as "RD" in table 4 on page 4, need to check these on the whole paper.
- The authors mentioned that they took the 152 retinal images of age group between 46 to 76 which had CAC scores <400 and >400, but you could see the other age group people below 46 also exists with CAC scores <400 as well as >400. So a better description explaining the reason for choosing the age group >46 is required to support their data.
- The authors mentioned that their proposed two settings which can fit specific applications such as clinical diagnosis and large scale retrieval, but brief description about these applications are missing here which reduces the reader's interest about this proposed technique.
- The authors need to represent either full representation of labels used in the tables or do mention separately in the theory explaining about the table, which is missing in most of the tables in the paper.
- Since some other papers claim that their training dataset is an unprecedented dataset, the authors need to explain the significance of their used data in the discussion section.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
