Multiple Fibers Per Voxel Detection by Fast Independent Component Analysis

Estimation of multiple fiber tracts within a voxel has been problematic in diffusion tensor imaging (DTI) tractography. After establishing from human measurements that DTI signals typically conform to nonGaussian statistics, a novel method was developed to resolve multiple nonGaussian components from DTI signals within single voxels using fast independent component analysis (ICA). The principal vectors of these components were then used to estimate the orientations of multiple fibers for tractography. Preliminary results, obtained after restricting the number of components to two in voxels identified as likely to contain multiple fibers, show many tracts that appear to be consistent with known anatomy but are not detected by conventional single-fiber tractography.