3D Reconstruction of Axonal fibers from Diffusion Tensor Imaging using Fiber Assignment by Continuous Tracking (FACT)

Introduction Diffusion-weighted MRI allows in vivo mapping of the diffusional properties of brain water and has revealed a high degree of diffusional anisotropy in white matter (1). Despite many studies of this anisotropy and the production of vector pictures showing fiber orientation within voxels of image planes, the actual reconstruction of neuronal projections by tracking of these vectors has never been accomplished. In the present study, we demonstrate an approach called Fiber Assignment by Continuous Tracking (FACT), which is able to achieve such 3D tracking of axonal projections (2). The results are subsequently validated by in situ tracking of known fiber tracts in rat and sheep brains. Materials and Methods MRI studies: Rat and sheep brains fixed in formalin were subjected to the study using 9.4 and 4.7 T GE Omega imagers, respectively. A diffusion-weighted 3D spin-echo sequence was used to record images that were diffusion-weighted along ten independent axes. An additional image with the least diffusion-weighted image was also recorded and six independent tensor elements were calculated for each pixel. A data size of 128 x 64 x 64 was acquired and zerofilled to obtain final resolution of 256 x 128 x 128. FOV was 32 x 16 x 16 mm (rat) or 90 x 60 x 54 mm (sheep). Truckina: One of the difficulties in the post-processing reconstruction of 3D fiber structures from diffusion tensor MRI data is related to decision-making for the connections between voxels in the image. This is illustrated in Fig. la, where the fibers (long curved arrows) are assumed to be confined to the 2D plane and the calculated fiber direction within each voxel is indicated by a straight open arrow. Starting from the voxel with an asterisk, tracking should follow the bold curved arrow. The most intuitive way to perform this tracking is by connecting each voxel to the adjacent one at which the fiber direction is pointing. However, when using this approach, the tracking (indicated by the dotted voxels) often deviates from the true fiber orientation, because the choice of direction is limited (26 in the case of 3D). This problem is avoided when tracking a continuous rather than a discrete number field (Fig. lb). Here, tracking is initiated from the center of a voxel and proceeds according to the vector direction. At the point where the track leaves the voxel and enters the next, its direction is changed to that of the neighbor. Due to the presence of continuous intercepts, this tracking now connects the correct voxels and the actual fiber (bold straight arrows) can be assigned. We therefore dubbed this approach FACT (Fiber Assignment by Continuous Tracking).