Principal eigenvector field segmentation for reproducible diffusion tensor tractography of white matter structures

The study was aimed to test the feasibility of utilizing an algorithmically determinable stable fiber mass (SFM) map obtained by an unsupervised principal eigenvector field segmentation (PEVFS) for automatic delineation of 18 white matter (WM) tracts: (1) corpus callosum (CC), (2) tapetum (TP), (3) inferior longitudinal fasciculus (ILF), (4) uncinate fasciculus (UNC), (5) inferior fronto-occipital fasciculus (IFO), (6) optic pathways (OP), (7) superior longitudinal fasciculus (SLF), (8) arcuate fasciculus (AF), (9) fornix (FX), (10) cingulum (CG), (11) anterior thalamic radiation (ATR), (12) superior thalamic radiation (STR), (13) posterior thalamic radiation (PTR), (14) corticospinal/corticopontine tract (CST/CPT), (15) medial lemniscus (ML), (16) superior cerebellar peduncle (SCP), (17) middle cerebellar peduncle (MCP) and (18) inferior cerebellar peduncle (ICP). Diffusion tensor imaging (DTI)-derived fractional anisotropy (FA) and the principal eigenvector field have been used to create the SFM consisting of a collection of linear voxel structures which are grouped together by color-coding them into seven natural classes to provide PEVFS signature segments which greatly facilitate the selection of regions of interest (ROIs) for fiber tractography using just a single mouse click, as compared with a manual drawing of ROIs in the classical approach. All the 18 fiber bundles have been successfully reconstructed, in all the subjects, using the single ROIs provided by the SFM approach, with their reproducibility characterized by the fact that the ROI selection is user independent. The essentially automatic PEVFS method is robust, efficient and compares favorably with the classical ROI methods for diffusion tensor tractography (DTT).