Segmentation Error for Large WMH lesion

[bianyueyan]

flair.nii.gz

Describe the bug

I used the antspynet.wmh_segmentation method for segmenting WMH of CSVD patients. It performed very well on the relative small WMH lesion, but not good for the large or very large WMH lesion. Especially for the interal part of large lesion, although it has very high intensity, the network gave a very low probability.

To reproduce

wmhprob = antspynet.wmh_segmentation(
flair=flair,
t1=t1,
do_preprocessing=True,
verbose=True
)

Expected behavior

Screenshots

ANTsPy installation (please complete the following information):

• Hardware [PC ]

• OS: [Windows ]

• System details [ ]

• Sub-system: [ ]

• ANTsPy version: [ latest version on 2025.09.10 ]

• Installation type: [ pip ]

Additional context
Add any other context about the problem here. Many issues are specific to particular data
so please include example data if possible.

[ntustison]

I wouldn't consider this a "bug", per se, but rather a case where this particular network performs poorly. Fortunately, there are multiple WMH segmentation algorithms that you can explore. In addition to the one you've already explored above, the following three have all been published and their weights made publicly available, which we ported to ANTsXNet (links to the corresponding examples):

• SYSU
• Hypermapp3r
• SHIVA

What I would suggest is that you look at all of them and see if one or more (or their combination) provides the desired result.