Simulation and Experimental Study of Multiple-Fibers Per Voxel Detection by ICA in DTI Tractography

The estimate of the eigenvector within a voxel for DTI tractography becomes incorrect when conventional single rank-2 tensor modeling is used in voxels likely to contain two or more intersecting fibers. This paper investigates the applicability of independent component analysis (ICA) to estimate individual tensors when two or three intersecting fibers are present within a voxel. After establishing non-Gaussianity of the diffusion tensor imaging (DTI) signals, a Monte Carlo simulation study was conducted to show the accuracy of recovering the orientations of two or three tensors (fibers) mixed within a voxel as a function of the crossing angles and the signal-to-noise ratio (SNR). Results suggest reasonable recovery of crossing angles above 20 deg at the typical SNR encountered in clinical 3T scanners for 25-gradient directions DTI data acquired in less than 8 min. Preliminary results of human brain tractography incorporating two or three ICA components in voxels likely to contain fiber-crossing show continuity in regions where single tensor tractography shows gaps attributed to fiber-crossing.