Jeong-Won Jeong PhD, Presenter: Nothing to Disclose

Harry T. Chugani MD, Abstract Co-Author: Nothing to Disclose

Diane C Chugani PhD, Abstract Co-Author: Research Consultant, Shire plc

Vijay Narayan Tiwari MD, PhD, Abstract Co-Author: Nothing to Disclose

To develop a new streamline tractography approach that can identify complex structures of multiple fibers crossing at voxels of clinical diffusion tensor images (DTI), we combined an independent component analysis (ICA) with a gaussian mixture model (GMM).

Whole brain DTI with 55 directional gradients at b=1000 s/mm2 and one b0 image was performed in 30 typically developing children (mean age: 10.0±3.3 years, 4.3-17.8 years, 21 males). For each voxel, a symmetric orthogonalization ICA was applied over a 33 neighborhood window in order to identify three fiber components, which are maximally independent in orientation. The eigen vectors of these components were then converted into corresponding Euler angles to initialize the orientatations of three gaussian tensors in the GMM. Quasi Newton Broyden-Fletcher-Goldfarb-Shannon method was used to optimize a set of GMM parameters (e.g., tensors and fractions of indivdial fibers). The resulting parameters were then used to recover the orientation and fractional anisotropy (FA) of individual fibers. For the comparison, we estimated multiple fibers using ICA alone and spherical deconvolution (SD). The streamline tractography was modified to accommodate multiple orientations in voxels by selecting the nearest orientation to interpolate onging fiber directions. Group percentage overlap maps of corticospinal tracts (CST) were compared to assess performance of each method.

Compared to other methods, the combination of ICA and GMM could isolate correct orientations of lateral projections (ROIs in Fig. a) and recover more consistent FA values in voxels where the CST intersects with the superior longitudinal fasciculus. The proposed method achieved higher correspondence with post-mortem histology,78 % voxel overlap (Fig. b), whereas other methods resulted in 67% (ICA) and 53% (SD) overlap.

This study demonstrated that individual tensors of multiple fibers crossing at a voxel can be recovered from clinical DTI data more reliably using ICA based GMM.

Our streamline tractography method combining ICA based GMM provided the most accurate tractography compared to post-mortem histology. This method can be used to study lateral projections of the CST th

Jeong, J, Chugani, H, Chugani, D, Tiwari, V, Independent Component Analysis-based Gaussian Mixture Model Tractography to Identify Intravoxel Crossing Fibers from Clinical Diffusion Tensor Images.  Radiological Society of North America 2011 Scientific Assembly and Annual Meeting, November 26 - December 2, 2011 ,Chicago IL. http://archive.rsna.org/2011/11012947.html