Acquiring optimal DWI data for tractography on post-mortem brain tissue

. To exclude spatial differences between the DWI datasets all data was acquired within the same scanning session. Additionally, to reduce geometric image distortions and to minimize susceptibility artefacts (e.g. due to air bubbles), a conventional diffusion weighted spin echo sequence (dwSE) was used: TE: 67.8 ms; TR: 2500 ms; FOV: 65x32 mm; matrix: 128x64; 5 slices; voxel size: 0.51x0.51x0.5 mm 3 . Diffusion parameters were: DELTA: 33.5 ms; delta: 27 ms, and seven different gradient strengths: 29, 44, 49, 56, 68, 80 and 111 mT/m. Each DWI dataset consisted of 3 non-dw and 61 dw (non-collinear directions) image volumes. A NEX=4 resulted in a SNR of 21. Two reconstruction methods were used for voxel-wise modelling of the underlying true fibre compositions. Method I: Spherical harmonics (SH) of order 0 th , 2 nd and 4 th were fitted to each voxel and a voxel-wise F-test for each SH was calculated (Alexander et al, 2002). For classification, user defined thresholds of the F-tests were used to categorize the voxels into one of the three orders of SH, representing isotropic, anisotropic or non-Gaussian diffusion profiles. The latter class includes mixed compositions such as crossing fibres. The F-test thresholds selected were: 1E-13 for segmenting iso- and anisotropic profiles and 1E-8 to segment anisotropic and non-Gaussian profiles. Since all DWI datasets had similar SNR the same thresholds were used. Method II: Persistent angular structure (PASMRI) is a q-space method that via a Fourier transformation searches for the best spherical fit to reflect the angular displacement density of water molecules. The peaks of the spherical function provide the estimated fibre orientations of a single or more fibres crossing within a voxel. Both methods are available in the Camino diffusion toolkit (Cook et al. 2006). Results Clusters containing non-Gaussian voxels become apparent for b-values > 2475 s/mm 2