Marieke van Rosmalen

Chapter 6 96 processing pipeline comprised data denoising, affine registration to correct for subject motion and eddy currents, b-spline registration to correct for echo-planar imaging distortions, tensor estimation using an iterative weighted linear least squares algorithm and whole volume fiber tractography (seed point resolution 1x1x1 mm 3 , step size 1 mm, seed FA threshold 0.15, FA track range 0.1-0.8, fiber length range 20-200 mm, angle threshold 15 o per step; Figure 6.1C ). This first automated processing step required approximately 35 minutes per data set to complete. Next, we manually defined slices with starting and ending points of tracts. Starting points were located next to the ganglion of nerve root C5, C6 and C7, ending points were located 5 slices further in the distal direction (5 minutes per data set). This aided a second algorithm to find all tract locations of the nerves using a tract density map (F igure 6.1D ) and specifies the appropriate region of interests (ROI’s) for nerve segmentation ( Figure 6.1E ). To pair ROI’s in the proximal starting and distal ending slices, the algorithm performs a connectivity analysis for all defined ROI’s. Every pair of ROIs with high connectivity is then defined as tract bundles which results in a reconstruction of the nerve roots ( Figure 1F ). Subsequently, the nerve root segments were constructed, using the predefined starting and ending slice ( Figure 6.1G ). These nerve root segments were used to standardize the site of extraction of diffusion parameters (FA, MD, AD, RD), i.e. next to the ganglion over a distance of 1 cm. This second automated part of the pipeline required approximately 5 minutes per data set to complete. Finally, we visually identified and labeled the selected tracts as the left and right nerve roots of C5, C6, or C7 (5 minutes per data set). If necessary, manual ROI’s were placed to optimize the result of the automated data processing (5 minutes per data set). When no tracts were found, nor with the algorithm, nor manually, the data set was excluded from further analysis. Finally, diffusion parameters per fiber tract were calculated using tract-based analysis. T2 mapping and T1 DIXON Dixon fat fraction maps were calculated using the water and fat image reconstructions of the vendor software. The data obtained with T2 mapping was processed using an extended phase graph fitting approach considering inhomogeneous B 1 +. 29,30 This method accounts for different T2 relaxation times for the water and fat component with the T2 of the fat component fixed to a value calibrated on the subcutaneous fat. Quantitative values of the T2 mapping and T1 DIXON were obtained using the same tract-based analysis used for DTI data. Data underwent registration to the same anatomical space (3D TSE SPIR image) as the DTI data. We obtained T2 relaxation time in milliseconds (ms) and fat fraction in percentages.