Probabilistic MRI tractography of the optic radiation using constrained spherical deconvolution: a feasibility study

Abstract

Background and purpose: Imaging the optic radiation (OR) is of considerable interest in studying diseases affecting the visual pathway and for pre-surgical planning of temporal lobe resections. The purpose of this study was to investigate the clinical feasibility of using probabilistic diffusion tractography based on constrained spherical deconvolution (CSD) to image the optic radiation. It was hypothesized that CSD would provide improved tracking of the OR compared with the widely used ball-and-stick model.

Methods: Diffusion weighted MRI (30 directions) was performed on twenty patients with no known visual deficits. Tractography was performed using probabilistic algorithms based on fiber orientation distribution models of local white matter trajectories. The performance of these algorithms was evaluated by comparing computational times and receiver operating characteristic results, and by correlation of anatomical landmark distances to dissection estimates.

Results: The results showed that it was consistently feasible to reconstruct individual optic radiations from clinically practical (4.5 minute acquisition) diffusion weighted imaging data sets using CSD. Tractography based on the CSD model resulted in significantly shorter computational times, improved receiver operating characteristic results, and shorter Meyer's loop to temporal pole distances (in closer agreement with dissection studies) when compared to the ball-and-stick based algorithm.

Conclusions: Accurate tractography of the optic radiation can be accomplished using diffusion MRI data collected within a clinically practical timeframe. CSD based tractography was faster, more accurate and had better correlation with known anatomical landmarks than ball-and-stick tractography.

Conflict of interest statement

Competing Interests: The authors have declared that no competing interests exist.

Figures

Fig 1. An example of the input data and ROIs used for fiber tracking.

The seeding/target ROIs are overlaid on registered anatomical images. (A-C) ROIs in the axial, sagittal and coronal planes include the LGN seed (blue), V1 target (red) and waypoint (green) areas, a frontal/midline exclusion zone (yellow) and an atlas-defined OR probability map (red/yellow) overlaid on the MNI152 standard brain. (D-F) Representative axial, sagittal and coronal slices of the ADC map. (G-I) Representative axial, sagittal, and coronal slices of an FA map. (J-L) FOD plots for each voxel and the optic radiation (yellow) overlaid on the FA map in the region of Meyer’s loop indicated by the cross-hairs in (G-I).

Fig 2. Complete CSD and B&S (probtrackX) tractography results for a single subject.

(A) Without using a threshold to remove voxels of low connectivity. The unthresholded CSD streamline reconstructions of the OR (yellow) in three planes are shown together with whole brain color-coded CSD tracks for visual reference. (B) The unthresholded B&S SDIs (green) overlaid on anatomical T1 weighted images in three planes. (C) Final thresholded CSD SDI. (D) Final thresholded B&S SDI. An optimal threshold (Table 1) was used to remove voxels of low connectivity such that the median FPR across all patients was 2.1%.

Fig 3. Left optic radiation SDIs for a single subject.

The OR of the same subject as in Fig. 2 is shown with no threshold (left), the optimal SDI threshold (middle) and 10 times the optimal SDI threshold (right). (A) SDIs based on the CSD model. (B) SDIs based on the B&S model.

Fig 4. Group results of optic radiation fiber tracking.

(A) The mean ROC curves with 95% confidence intervals (dashed lines) for the CSD and B&S based tractography. A line of unity is shown for comparison. (B) Box plot of the AUCs for the CSD and B&S tractography algorithms. The CSD algorithm resulted in a significantly (p

Fig 5. Dependence of OR tractography results on the number of diffusion directions in a single subject.

Tractography was performed on a single subject from the human connectome project (www.humanconnectomeproject.org). The CSD tractography was performed on pre-processed 90 direction diffusion data, with a b-value of 3000 and 1.25mm isotropic voxel resolution, subsampled between 20 and 88 directions. The sensitivity, specificity and similarity indices were all computed using the fully sampled DWI data as the gold standard. All MLA-TP distances were within one standard deviation (dashed lines) of the median dissection distance (solid line).