Reinforcement Learning Based Vocal Fold Localization in Preoperative Neck CT for Injection Laryngoplasty

Round 1

Reviewer 1 Report

This paper proposed a reinforcement learning algorithm based on a mirror environment for localizing the right and left vocal folds in preoperative neck CT. This mirror environment allows for localizing both the folds by a single training session which treats the left fold as a flipped version of the right fold. Following are some comments to improve this communication:

1. In Line 44, the definition of the Cartesian axes is suggested to be also presented in a figure for better illustration.

2. In the paragraph after Line 90, the value function and discounted cumulative reward should be explained in detail, including the parameter definition, purpose of building these functions and how to use them, etc.

3. In Line 92, how do you calculate the policy ratio?

4. Using the mirror environment, does the single training perform the same for the left and right localizing mission? The accuracy difference between localizing the left and right vocal folds is suggested to be compared and analyzed.

5. In Line 124, 1e-4 should be changed to scientific notation.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Interesting manuscript. In practice, the identification of the vocal cords for augmentation injections is performed under endoscopic control. In this way, the amount of substance to be injected can also be checked. But why not another method...

Author Response

Please see the attachment.

Author Response File: Author Response.pdf