All regions of interest (ROIs; LGN, V1, V2, V3, V4, and V5) were detected using Juelich probabilistic histologic atlas
19: First of all, anatomical T1-weighted 3D volumes were nonlinearly coregistered to preprocessed diffusion as previously described.
20 The New Segment Option of SPM8 tool was used to drive the registration procedure comparing cerebro-spinal fluid (CSF) probability maps extracted from T1 and b0 volumes. Usually, an affine mapping of T1s to fractional anisotropy (FA) maps is performed to coregister structural scans to diffusion data. This approach suffers from two inherent flaws: (1) some nonlinear local geometric distortions still persist on diffusion data even after preprocessing, and (2) FA and T1 maps provide different contrasts in brain tissues, because the former is mostly focused on white matter, while the latter mostly highlights gray matter structures. The use of a nonlinear procedure can reduce misregistration by providing a mapping, which more closely follows the anatomy in diffusion space. On the other side, CSF can be extracted with a rather good accuracy both from diffusion data and T1s; thus, spatial priors that are common to both diffusion and T1 spaces can drive CSF-based registration. We know that partial volume effects could affect CSF; for this reason, we used CSF probability maps instead of crude CSF binary ones for warping T1s to diffusion images. This choice allowed performing a weighted registration by further taking into account potential partial volume inaccuracies that might appear in both maps. Coregistered T1s were then normalized to Montreal Neurological Institute (MNI) stereotactic space by means of FSL utilities FLIRT and FNIRT (in the public domain,
http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/). Estimated warping fields were later on inverted and applied to Juelich template ROIs in order to have them represented in native space of our subjects. The 50% atlas was used (i.e., only areas that belonged to the claimed structure with a probability ≥50% were used). These ROIs were eventually visually inspected by a radiologist with 20 years of experience (MG) and refined to prevent possible misalignments; 3 of 16 subjects showed no more than two spurious voxels in their LGN ROIs (two subjects for left LGN and one subject for the right one), which were erased. Regions of interest correspondent to right and left LGNs were used as seeds for the probabilistic CSD based tractography; all tractographic reconstructions were obtained using MRtrix software package (in the public domain,
http://jdtournier.github.io/mrtrix-0.2/index.html).
21 The following reconstruction parameters were used: maximal spherical harmonics degree of 8, maximum fiber length 150 mm, step size 0.2 mm, minimal fiber orientation distribution function (fODF) amplitude 0.15. The latter parameter constitutes a conservative approach that might cause potential underestimation; however, we preferred to use it in order to keep to a minimum false positive tracts,
22 because in this way only voxels with a high probability to belong to white matter are involved in the tracking procedure. Target ROIs were moderately dilated to include gray/white matter boundaries; in this way we ensured that streamlines were able to reach target ROIs for subsequent connectivity analysis. This step could cause an overlapping between different visual area-ROIs (V-ROIs); assigning to a V-ROI those regions that had in the neighborhood, on average, the highest probability of accordance with a specific visual area of the histologic atlas solved these conflicts. All tracts were automatically colored, according to streamlines directions, in green (anterior–posterior direction), blue (cranial–caudal direction), and red (left–right direction). From each LGN, 200,000 streamlines were generated.