MR tractography holds the promise of deriving neuroanatomical information from non-invasive MR studies, but its performance in practice is usually difficult to verify. Going beyond the simplest DTI tensor model is probably essential. Probabilistic tractography has been used for a number of years, but protocols for use in clinical scanners must still be optimized on a tract-by-tract basis. In the current study we aimed to develop an appropriate protocol for defining the cortical spinal tract (CST) for clinical examination.

Twenty-five healthy adult subjects (13 females), with an age range 20-83 years (mean 38), were recruited, and underwent MR scanning using a 1.5 Tesla GE Signa system. A T1-weighted axial volumetric image was acquired using the FSPGR sequence (25.6 cm² FOV; 1 mm isotropic voxels) and axial diffusion-weighted images using a single-shot SE-EPI sequence with FOV 32x32 cm, 3 mm slice-thickness, in-plane resolution 128x128, b-value = 900 s mm^-2, and 64 diffusion-weighted directions + 7 unweighted scans. The protocol was approved by the local Ethics Committee and written informed consent was obtained from all participants. Cortical parcellation/labelling of the volumetric images was performed by Freesurfer¹. Volumetric images and associated labels were aligned to DWI volumes using FSL/FLIRT, followed by nonlinear deformation². Basic DWI processing was performed using the FMRIB software library (http://www.fmrib.ox.ac.uk/fsl), applying bedpostx and probtrackx for probabilistic tractography³. The starting point of the CST (seedmask) was identified using voxels labelled by Freesurfer as lying within the precentral gyrus, separately for left and right sides. In addition fibres were constrained to pass through the posterior limb of the internal capsule (PLIC) (waypoint) and thence to the pons (termination mask). The result of the tractography was a voxelwise estimate for CST connectivity (fibre count). All subject data was linearly aligned to the FMRIB58 FA template. Voxels that contained average fibre numbers of at least 0.4% maximum were used to define a consensus CST mask. For each side of the brain, fibre count images were resampled to yield exactly 100 slices in the z-direction to cover the distance from the pons to the upper limit of the precentral gyrus. To identify protocol parameters that might improve tract definition, two operational variables were defined as quality metrics for each subject:

i) False negative rate (FNR) – false negative voxels within the seedmask without emergent fibres / all seedmask voxels;

ii) False discovery rate (FDR) – false positive voxels containing fibres but lying outside the consensus mask / all fibre-containing voxels.

Two methods of thresholding the fibre count, to reduce FDR, were tested – per subject thresholding, and slicewise thresholding, considering that the main course of the CST runs superior-inferior from the motor cortex to the pons. Several normalisation parameters were considered, and the best threshold level was side to minimize FNR and FDR.

Slicewise total fibre numbers are shown in Figure 1. The pattern is qualitatively similar for all subjects: numbers increases from the highest slice reaching a peak around the level of the PLIC, before falling to a low value as the tract descends to the midbrain, with the voxel maximum fibre number showing a roughly inverse pattern, conditioned by the relative number of fibres originating at or above a given level. A performance curve for different threshold values is shown in Figure 2. The expected FNR/FDR trade-off is observed, starting from mean FNR 0.30 ± s.d. 0.13 and FDR 0.37 ± 0.13 for threshold=0. Slicewise thresholding is slightly better. False positive/negative results are demonstrated in Figure 3 for a typical subject.Methods for performing tractography of the CST have recently been examined⁴, and no single method was found to be definitely superior. Probabilistic tractography naturally generates streamlines that are unlikely to correspond to neuronal fibres, and so a method for culling false fibres is needed. Constraining fibre paths to pass through the PLIC and pons produces an adequate performance if combined with a fibre number threshold. The number of fibres in each slice (Figure 1) is determined mainly by the number of seed voxels above, and the compactness of the fibre tract (demonstrated by the maximum fibre number). Allowing for this slicewise variation marginally improves tract delineation, especially with a high threshold (Figure 3), and is thus useful for defining the main part of the tract below the corona radiata. At and above this level, additional methods will be required, especially in the lateral portions of the corona radiate and centrum semiovale, which lie on the same level as the more medial tract and are readily culled with a slicewise threshold.