Anatomically informed multi-level fiber tractography for targeted virtual dissection

Abstract

Objectives: Diffusion-weighted MRI can assist preoperative planning by reconstructing the trajectory of eloquent fiber pathways, such as the corticospinal tract (CST). However, accurate reconstruction of the full extent of the CST remains challenging with existing tractography methods. We suggest a novel tractography algorithm exploiting unused fiber orientations to produce more complete and reliable results.

Methods: Our novel approach, referred to as multi-level fiber tractography (MLFT), reconstructs fiber pathways by progressively considering previously unused fiber orientations at multiple levels of tract propagation. Anatomical priors are used to minimize the number of false-positive pathways. The MLFT method was evaluated on synthetic data and in vivo data by reconstructing the CST while compared to conventional tractography approaches.

Results: The radial extent of MLFT reconstructions is comparable to that of probabilistic reconstruction: [Formula: see text] for the left and [Formula: see text] for the right hemisphere according to Wilcoxon test, while achieving significantly higher topography preservation compared to probabilistic tractography: [Formula: see text].

Discussion: MLFT provides a novel way to reconstruct fiber pathways by adding the capability of including branching pathways in fiber tractography. Thanks to its robustness, feasible reconstruction extent and topography preservation, our approach may assist in clinical practice as well as in virtual dissection studies.

Keywords: Corticospinal tract; Diffusion MRI; White matter.

Fig. 1

(Left) Current fiber tractography methods such as deterministic FOD-based tractography reconstruct only a subset of the pathways (blue). However, by propagating along the FOD orientations that were not used by a conventional tractography algorithm, the reconstruction can be iteratively extended by adding new sets of branches per iteration (red and green) leading to a final tractography result consisting of multiple levels. (Right) The pipeline of the algorithm. a The tract produced by the deterministic CSD-based tractography includes points with multiple FOD peaks, some of which are ignored. b Using these points as seeds with the unused peaks as initial locations, another iteration of CSD-based tracking is performed to obtain a new level of the result. c In the last stage only the tracts that enter the pre-defined target region are retained. The background picture on the left of the whole-brain fiber tractography result is taken from [33] with permission

Fig. 2

Performance of the considered methods in phantoms (FA map). The top row shows the results of MLFT and the bottom row those of the iFOD2 algorithm. The middle row illustrates the target fibers per column (orientation-based colored FA map). The same single seed point (yellow sphere) was used for both algorithms. The results of iFOD2 were subsampled for easier visual assessment. Streamlines' colors are based on orientation color-coding

Fig. 3

Tracts reconstructed by MLFT on the phantom data (FA map) at multiple SNR levels. Considerable angular errors are only observed at SNR = 15: increased number of branching configurations and direction perturbations up to 30º (red arrows). At SNR = 25, there is a minor angular deviation below 10º (red arrow). Streamlines are colored using standard orientation color-coding

Fig. 4

Fiber pathways reconstructed by the deterministic CSD-based approach (left) and MLFT with two (middle) and three (right) levels from the same seed region (green) with the same target region (yellow, the motor cortex) using MASSIVE dataset. Adding the second-level branches (red) to the pathways obtained at the first level (blue) improves the extent of the reconstructed bundle. Using three-level reconstruction from the same seed region does not show coverage improvements over the two-level reconstruction

Fig. 5

Corticospinal pathways reconstructed by the conventional deterministic CSD-based tractography, MLFT, iFOD2 and global tractography using the MASSIVE data. The motor cortex is shown in yellow. Some of the pathways reconstructed by iFOD2 enter the motor cortex and diverge into the CC propagating into the contralateral hemisphere

Fig. 6

The CST reconstructions obtained by MLFT, iFOD2, GT and CSD-based tractography using the HCP data. The reconstructions by MLFT bundles are in line with the observations in Fig. 5 and consistent with each other. iFOD2 also achieves high motor cortex coverage. The extents of the GT-reconstructed bundles are comparable to the ones obtained by MLFT, but with less satisfactorily spatial coverage

Fig. 7

Radial extents of the reconstructed CST bundles for both left and right hemispheres. MLFT (blue) is shown to improve the radial extent compared to the conventional deterministic CSD-based tractography (green). iFOD2 (orange) and MLFT (blue) appear to have comparable radial extents. GT (red) achieves high radial extent on the MASSIVE dataset, while on HCP data the extent is primarily low

Fig. 8

Density of the motor cortex coverage by the reconstructed CST bundles considering with angular coordinate starting at tempo-lateral point of coronal projection of the motor cortex and increasing towards superior motor cortex separately for each hemisphere. All the algorithms appear to densely cover superior part of the motor cortex. However, MLFT (blue) consistently covers most lateral part of the motor cortex with its density more evenly distributed compared to iFOD2 (orange). GT (red) also occasionally covers temporo-lateral motor cortex, although the coverage is very sparse. CSD-based tractography (green) primarily covers superior motor cortex part in all the subjects

Fig. 9

Coronal and sagittal views of the left CST reconstructed by MLFT and iFOD2 using the MASSIVE data. The fiber pathways are colored according to the locations of their endpoints in the motor cortex. The pathways reconstructed by MLFT are shown to have a clearer topographic organization

Fig. 10

Normalized distributions of the distances from each pathway to the nearest neighbor based on the MADF distance for all the processed subjects. The distributions of the distances appear to be similar across subjects