search

josedolz / LiviaNET

This repository contains the code of LiviaNET, a 3D fully convolutional neural network that was employed in our work: "3D fully convolutional networks for subcortical segmentation in MRI: A large-scale study"
MIT License
161 stars 52 forks source link

About the MHD #17

Closed gangli95 closed 6 years ago

gangli95 commented 6 years ago

Hi,Jose. Recently, I counted the Dice and HD, and Dices can reach 0.9, but HD is around 3 or 4, which is so different with your result. What are the differences between HD and MHD, and is there anything I need to do when I count HD?

josedolz commented 6 years ago

Hi @gangli95

The standard HD takes into account the distances from all the points in both contours. On the other hand, the modified HD takes only the 95% of the computed distances, putting aside some outliers. That is why the values of MHD are typically smaller than HD values.

gangli95 commented 6 years ago

Hi,Jose image

I found this equation in Dubuisson's paper, did you takes the 95% of the computed distances and then count the mean of the 95% of the computed distances?

josedolz commented 6 years ago

Hi @gangli95

sorry for my late answer, I have been quite busy the last months.

Here is a piece of code to compute the MHD (evaluation.py):

https://github.com/hjkuijf/MRBrainS18

gangli95 commented 6 years ago

Hi, @Jose It's so kind of you to give me this reply, and it will never be late for the answers. I'll read the code carefully. Thanks very much!