Artikel
White matter tract reconstruction using semiautomatic seed voxel generation
Suche in Medline nach
Autoren
Veröffentlicht: | 4. Juni 2012 |
---|
Gliederung
Text
Objective: Diffusion Tensor Imaging and fiber tractography became a popular method for reconstruction of major white matter tracts in the human brain in-vivo. Up to now several approaches are published dealing with the calculation of localization and spatial extent of fiber bundles. To achieve maximum safe tumor volume resection, knowledge about subcortical functional structures is important. Now a new approach is presented to overcome the problem of seed voxel generation for user-independent fiber tracking.
Methods: After calculation of one initial fiber tracking with a line propagation algorithm and restriction of the result with include/exclude regions, the centerline of the segmented anatomical structure is calculated. Different planes are calculated along the centerline upright to its direction, determining new seed regions for tracking. This new set of seed voxels is then used for multiple fiber tracking. Summed up results of fiber tracking finally represent levels of tract membership, with voxels included in many reconstructions are likely to belong to the structure of interest.
Results: Evaluation is done using an anatomical software-phantom with given ground truth and different noise levels. Quality of reconstruction is evaluated with a spatial overlap measure called Dice Similarity Coefficient (DSC). In case of the modelled corticospinal tract initial fiber tracking achieved a DSC of 58.9%. For 128 automatically calculated new seed regions or subgroups of seed regions, the DSC and thereby tracking quality could be increased to a maximum DSC of 92% for 64 seed regions with a seed region offset of 3 mm (average DSC: 83.57% ± 6.35%). Standard fiber tracking (including restrictions) took about 10 s. The advanced approach took up to 2 min, which is still practicable for clinical routine.
Conclusions: An approach for increasing the feasibility of standard fiber tracking results with semiautomatic seed voxel generation has been presented and evaluated, showing an increasing overlap between ground truth and advanced processing result. Furthermore for intraoperative visualization different kinds of feasibility levels could be modelled, by taking different levels of memberships for visualization.