How you collect datum from two radiologists and how to use to get consistency loss ?

[PyZheng18]

you remind in paper that there are two radiologists in your test. I wonder if the two radiologists look the same X-ray images or different? if they look the same images, how do you use the one X-ray image with two annotations to caculate the attention consistency loss with uncertainty?

[JamesQFreeman]

In fact, there are multiple solutions:

1. duplicate the image, then you have (im1,gaze1) and (im1,gaze2) as two individual item;
2. chose one and discard the other one, which means you have (im1,gaze1)

You should choose your approach by your image. We have test a few: knee x-ray, mammography, chest x-ray and CatVsDog. Viewer variance on knee-xray & catvsdog are smaller than mammography and chest x-ray. Approach_2 can be applied to smaller variance data while approach_1 can be applied to both.

[PyZheng18]

Thanks! I want to follow your work. Unfortunately, i have no idea to get the annotations dataset of radiologists. So would you mind publishing or sharing the dataset used in your paper?