95% Hausdorff distance 0

Support
1

Hi there,

I’m working on an imaging fusion project that combines a small-FOV image with finer details and a large-FOV image with better description of surrounding anatomical structures.

To evaluate the performance ( reproducibility), I did the following:

  1. Add two scalar volumes after a rigid transform of one volume while keeping one fixed to get a fused image(volume)
  2. Intensity based segmentation A of the fused image(volume)
  3. Repeat step 1-2 to get the second segmentation B.
  4. Calculate the image metric with Hausdorff and DICE.

Here is the question. I noticed that both mean and maximum of Hausdorff distance values are reasonable but 95% is 0. I wonder if it’s something I am doing is not correct or that’s what I should expect? I came across a previous post mentioning the 95% could be 0.

Thank you for your time and help in advance.

1000016459
10000164591080×2400 398 KB

1000016458
10000164581130×1550 176 KB

2

If HD 95% is 0 then it means that 95% or more surface points have 0.0mm (or less than one voxel) error. This is possible if you use synthetic data and the alignment is nearly perfect.

3

@lassoan Thank you for the reply!