Grey–white matter boundary definition

[Yingying-H]

I used LayNii + segmentator for segmentation (upsampling our dataset into 0.2mm iso). I am trying to expand the grey matter as much as possible, but I am unsure how far the GM boundary should extend. Should I include all bright regions (as seen in the lower-right image), or only the middle part of the bright band (as I have currently done)?

The top left is one participant's anatomy brain (0.4mm iso), the top right is the gradient magnitude of the filtered image (upsampled anatomy with 0.2mm iso of the same participant's data), and bottom two images show the same regions with segmentation overlays (red = GM, yellow = WM).

[ofgulban]

Thanks for your question @Yingying-H . It is hard to give a definite answer for 3D image segmentation quality based on a single slice. However, assuming that this slice is representative of the rest of the segmentation, the lower left panel looks pretty good to me. With regards to the gradient magnitude images, if the boundary is cutting the brightest voxels in gradient magnitude images, I think this is a good point to believe that it is the most likely location of the boundary. This is similar to looking at where the zero crossing is in the second spatial derivative. Feel free to ask further if there is something not clear.