A Fast Method for Whole Liver- and Colorectal Liver Metastasis Segmentations from MRI Using 3D FCNN Networks

Round 1

Reviewer 1 Report

The purpose of the study was to develop a deep learning model trained on an in-house dataset of 84 MRI volumes to rapidly provide fully automated whole liver and liver lesions segmentation from volumetric MRI series. A cascade approach was utilized to address the problem of class imbalance. 

The idea was better to explore but it doesn't provide enough and comprehensive detail. Much is needed to improve the paper as follow;

• Highlight the significance of the approach
• No analysis is given to verify the results.
• No algorithm is provided
• No comparison is given with the previous results to see the achievements of the current study.
• No experimental results are given.

Minor Issue

• English needs to be improved
• Math equations are given using sentences rather than using math editor
• References are not enough

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

This paper has addressed an interesting topic. However, needs minor revisions.

1. A literature review is missing. Add one new heading name as a literature review.
2. Increase the introduction part.
3. Write the algorithm in the method section
4. A confusion matrix is missing
5. The discussion part needs to be elaborative.
6. The activation maps of the proposed model are missing
7. Main equations of CNN in missing
8. Enhance the comparative study, apply the other variant of CNNs
9. The need of 3D CNN is not clear.
10. Cite the following papers 

a) Nie, D., Cao, X., Gao, Y., Wang, L., & Shen, D. (2016). Estimating CT image from MRI data using 3D fully convolutional networks. In Deep Learning and Data Labeling for Medical Applications (pp. 170-178). Springer, Cham.

b) Sun, M., Lu, L., Hameed, I. A., Kulseng, C. P. S., & Gjesdal, K. I. (2021). Detecting Small Anatomical Structures in 3D Knee MRI Segmentation by Fully Convolutional Networks. Applied Sciences, 12(1), 283.

c) Roy, S. S., Rodrigues, N., & Taguchi, Y. (2020). Incremental dilations using CNN for brain tumor classification. Applied Sciences, 10(14), 4915.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

In this study, a deep learning model was developed to be trained on an in-house dataset of 84 MRI volumes for rapid and automatic whole liver and liver lesions segmentations from volumetric MRI series. The authors used a cascade approach to address class imbalance. The trained model achieved an average Dice score for whole liver segmentation of 0.944∓0.009 and 0.780∓0.119 for liver lesion segmentation. The authors also demonstrated that applying this method to a not-annotated dataset could create a complete 3D segmentation in less than 6 seconds per MRI volume, with a mean segmentation Dice score of 0.994∓0.003 for the liver and 0.709∓0.171 for tumors compared to manual corrections applied after the inference was achieved. The authors finally concluded that the availability and integration of their method in clinical practice may improve diagnosis and treatment planning in patients with colorectal liver metastasis and open new possibilities for research into liver tumors.

The manuscript was written in a clear and logical manner. I only have a few comments for the authors to address before it can be considered for publication.

1. Could the authors provide any histograms to show the class distribution of the training and test datasets?
2. How did the authors split the training and test sets?
3. Did the authors use transfer learning before training the models, or just did the authors train the model from randomly initialized weights? 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Author revised the paper as per the comments