Arcuate fasciculus delineation by means of diffusion compartment imaging based tractography

Purpose: The arcuate fasciculus (AF) is a crucial language pathway in the human brain connecting Broca's area in the frontal lobe to Wernicke's area in the temporal lobe. Using tractography to analyze the AF, investigators have identified anomalies in a variety of diseases1. Currently, white matter fiber tractography is most commonly implemented using the principal diffusion direction of the diffusion tensor (DT)2. At the resolution of diffusion weighted imaging (DWI) acquisitions, many voxels contain multiple fibers crossing, branching and fanning which infringe the assumption of a single fiber population per voxel imposed by the DT. Thus tractography algorithms based on the DT can follow false tracts or can prematurely stop due to isotropic tensors or to major orientation inconsistencies between neighboring voxels. For instance, tractography of the AF based on a DT model, due to crossing with the superior longitudinal fasciculus III (SLFIII) in the pars opercularis, terminates prematurely instead of extending anteriorly to the pars triangularis. In this work, we evaluated the ability of diffusion compartment imaging (DCI)3 based tractography to recover the complete length of the AF and compared it with that achieved by DT tractography. Methods: Fifteen healthy volunteers (nine male; mean age 9.1 years; SD 3.3; range 5-15) were imaged on a Siements 3T Skyra scanner using a 64 channel head coil. The study was approved by the institutional review board (IRB) and all participants provided informed consent. The MRI protocol included the acquisition of a structural T1w MPRAGE sequence and a DWI acquisition consisting of a CUSP90 sequence3 which includes 30 diffusion-encoding gradients on a shell at b=1000s/mm2 and 60 extra gradients in the enclosing cube of constant TE with b-values up to 3000s/mm2. Eddy current distortion was minimized using a twice-refocused spin echo sequence. The DWI were aligned to the T1w MRI with rigid registration (using the mean b=0 image as moving image) and the gradients were reoriented appropriately. The DT model was estimated using a robust least squares algorithm4 . The DCI model was estimated following Scherrer et al.5. Finally,