Hello everyone,
I am currently working on hydrocephalus . I need do segment the lobes of the brain in MRI of children with hydrocephalus . I saw a few extensions on 3D slicer and some sites like VolBrain but they are only able to segment in near normal appearing brain. I am ok with semi automatic segmentation also but I would prefer if I can do it in approx 20-30 minutes per scan. Does anyone have any suggestions ?
Thanks in advance
Regards,
Preetham S.D.
Fellow in pediatric neurosurgery, Amrita institute, Kochi,India
Hello. Do you need the true anatomical lobes (as meshes) or regional volumes (i.e., as a value)? If ventricular enlargement isn’t too severe I would try TotalSegmentator or NNInteractive (GitHub - coendevente/SlicerNNInteractive: A 3D Slicer extension for efficient segmentation with nnInteractive. · GitHub). There are also other MONAI/n-Unet based methods.
Thanks for replying Thomas
No I don’t need true anatomical lobes . If I can segment cerebrum, deep nuclei, brainstem and cerebellum. It would be fine
I tried both of them . The hydrocephalus I’m working on are quite severe so it didn’t work. Also I am working on CISS/Fiesta sequences .
Ah, I see. Most brain segmentation packages are not trained on CISS/FIESTA sequences, so that may be a challenge. I would suggest deformable registration of one known segmented volume to your target followed by manual correction:
- Obtain a (‘normal’) pediatric template with lobar labels.
- Register using ANTs.
- Import labels into 3D Slicer.
- Manually correct the boundaries.
Even when the registration is imperfect, correcting the propagated lobe map may be much faster than drawing the lobes from scratch… I have not done this, so this method may not work well, just a suggestion…
Thanks for the advice Thomas. I will try it once and let you know how it goes