All,

I am attempting multi-channel segmentation of brains with a variety of stroke lesions. While segmentation with mfast seems painless, the initial BET brain extraction is problematic.

BET does quite well at extracting our T2-weighted images at it's default settings. The T1-weighted images, however, do not extract cleanly. I have tried numerous iterations, adjusting the fractional intensity threshold, using -R, -S, or -A2, and all result in erroneously included and excluded voxels. That is, an unacceptable amount of grey matter is removed and a great deal of bone, muscle, and other tissue remains. In addition, large swaths of more shallow lesions are excluded, defeating the purpose of the process.

The T1 and T2 images have already been transformed to the same atlas space. Given that, along with the quality of our T2 BET extraction, does the following solution sound reasonable?

- Use 'bet -m' to extract the T2 and generate a binary mask
- Use the mri_mask tool from Freesurfer to apply the T2 mask to the T1 image
- Proceed with mfast using the extracted T2 and the masked T1

Alternatively, if this process sounds inappropriate, does anyone have suggestions for better T1 extraction?

Thank you for any help you can provide.


William E. Janes
OTD Student
Washington University School of Medicine
Program in Occupational Therapy
Campus Box 8505
4444 Forest Park Avenue
St. Louis, MO 63108
(618) 973-5344