Review Reports

Round 1

Reviewer 1 Report

This manuscript presented a more efficient/accurate method to segment liver cyst.  The result was quite encouraging.  I just have a few questions.

1. The image data were taken from the Polycystic Kidney Disease Repositor.  However, there was no mention of which magnetic fields those images were taken from, as magnetic field would have strong impact on image signal contrast.   In other words, would signal to noise ratio an issue for these segmentations?  Also there was no mention about the number of images were evaluated in this study.
2. Since many abdominal images suffered from motion artifact, does this technique only work well with perfect stilled images?
3. Some MRI images suffered inhomogeneous signal intensity through out the liver region.  Would it be a problem for this algorithm ?

Author Response

Response to review 1 comments

This manuscript presented a more efficient/accurate method to segment liver cyst.  The result was quite encouraging.  I just have a few questions.

Thank you for this positive comment.

 

Point 1

The image data were taken from the Polycystic Kidney Disease Repositor.  However, there was no mention of which magnetic fields those images were taken from, as magnetic field would have strong impact on image signal contrast.  In other words, would signal to noise ratio an issue for these segmentations?  Also there was no mention about the number of images were evaluated in this study.

Response:

Thank you for pointing out this omission. We now have specified the field strength as 1.5T (n=5) and 3T (n=12). The algorithm IRIS works well at both field strength, and 72 images on average per liver imaging volume. In the revision, we have added comparison between 1.5T and 3T, see the new first paragraph in Results.

Fundamentally, noisy images may be difficult to segment, and there are many effective denoising techniques, including latest deep learning ones. In our experience, ADPKD liver cysts are very bright with similar CNR on T2 weighted images acquired at both 1.5T and 3T, see the new first paragraph in Results.

 

Point 2

Since many abdominal images suffered from motion artifact, does this technique only work well with perfect stilled images?

Response:

Thank you for pointing out this motion artifact issue. Motion artifacts may not be tolerated by human observer radiologists in clinical practice and would affect IRIS software’s performance. In this work, T2 weighted images were acquired using breath holding, but there may be potential slice misregistration that affect cyst volume measurements. Motion artifacts may be regarded as degrading conspicuity, and we have modified discussion on limitation by adding:

A cause of poor conspicuity is motion artifacts, which should be minimized as required so in clinical practice.

Higher-resolution volumetric imaging with respiratory motion compensation(17-24) may help with accurate liver cyst volume measurement and overcome the potential slice misregistration from multiple breath holds that affect cyst volume measurements.

 

Point 3

Some MRI images suffered inhomogeneous signal intensity through out the liver region.  Would it be a problem for this algorithm?

Response:

Thank you for pointing out this issue of signal intensity inhomogeneity, which is a main cause of failure for automated algorithm. We now clarify that the IRIS tool is developed to address this problem. We have now modified the 2^nd paragraph in the discussion by inserting the following comments:

A major cause of failure for automated segmentation of numerous lesions in ADPKD is the signal intensity variation across the liver volume. The interactive features in IRIS address this unreliability by clicking into a small region where intensity variation within a cyst is small and segmentation can be robustly and rapidly performed using one click as we have learned from segmenting the bright left ventricle from the surrounding dark myocardium in cardiac MRI(9, 10).

Author Response File: Author Response.docx

Reviewer 2 Report

The introduction is focused on the topic, but in order to increase the strength of the paper, I would add some sentences about if there are other segmentation programs available for this specific tool, therefore I would move some sentences from the discussion to the introduction.   All the figures and tables should contain the Legends for the Acronyms used.

Author Response

Response to review 2 comments

We thank reviewer 2 for the helpful comments, which are partitioned into the following two points.

 

Point 1:

The introduction is focused on the topic, but in order to increase the strength of the paper, I would add some sentences about if there are other segmentation programs available for this specific tool, therefore I would move some sentences from the discussion to the introduction.  

Response:

Thank you for the helpful suggestion. We have moved discussion on other interactive techniques into the introduction: Manual segmentation can be improved with various interactive segmentation methods that use various user inputs, such as partial segmentation, as initialization to an automated output; however, these techniques are all restrictive and tedious(4).

 

 

Point 2:

All the figures and tables should contain the Legends for the Acronyms used.

Response:

The Legends for the Acronyms used are now provided.

Author Response File: Author Response.docx