Yuan Liu1,2, Benjamin Odry2, Hasan Ertan Cetingul2, and Mariappan Nadar2
1Vanderbilt Institute in Surgery and Engineering, Vanderbilt University, Nashville, TN, United States, 2Medical Imaging Technologies, Siemens Healthcare, Princeton, NJ, United States
Automatic
brain extraction, as a standard pre-processing step, typically suffers from a long
runtime and inaccuracies caused by brain variations and limited qualities of MR
images. We propose a generic supervised learning framework that builds binary
classifiers to identify brain and non-brain tissues at different resolution
levels, hierarchically performs voxel-wise classifications for a test subject,
and refines the brain boundary using narrow-band level set technique on the
classification map. The proposed method is evaluated on multiple datasets with
different acquisition sequences and scanner types using uni- or multi-contrast
images and shown to be fast, accurate, and robust.
