Summary

This chapter presented a fully automated hybrid intracranial boundary detection algorithm that has proven effective on clinical and research MR data sets acquired from several different scanners using PD-T2 spin-echo sequences. The algorithm has three sequential steps that provide first background removal, then an initial brain mask, and then a final brain mask. In the first step, the head is localized using histogram analysis, and a region that completely surrounds the brain is generated. The second step applies a nonlinear anisotropic diffusion filter and sets an automated threshold to produce a mask that isolates the brain within the head region of the first step. The third step uses this mask as an initial position for an active contour algorithm to determine the final intracranial boundary. This algorithm was robust in the presence of RF inhomogeneities and partial volume effects. It is in regular use for studies of multiple sclerosis lesions and MRI-PET registration studies. The chapter also surveyed several other methods for segmenting the brain from the head, including T1 MR images.

0 0

Post a comment