Question about Average Surface Distance (ASD)

[zzzqzhou]

Dear @liuquande ,

I tried to reproduce the result in your paper. However, there are some question about the metric ASD. Since I can't find any codes in this repository that calculate the ASD on test images. In "train.py" function "test", you only calculated the Dice score. So I added the ASD function from "medpy.metric.binary.asd" to calculate the ASD score. But, I can't reproduce the ASD result and the Dice score is close to your paper's results.

I found that using ndarray of different dimensions as input to "binary.asd" will get very different results. For example, I used a 3-dim ndarray shaped as [1, 384, 384] as input, the ASD is 1.08 mm, and then I changed the ndarray into a 2-dim ndarray shaped as [384, 384], the ASD result became 5.45mm, which had a big gap compared to the results in your paper.

I hope you can illustrate how you calculate the ASD in your method. Also, it is much better that you can add the code of calculating the ASD score.

Best.