gms | German Medical Science

Objective: Diffusion Tensor Imaging (DTI) and fiber tractography are popular and routinely used techniques to reconstruct and visualize major white matter tracts in the human brain in-vivo. During neurosurgical procedures, reliable and precise information about the spatial location, its course and spatial extent is essential to minimize postoperative deficits while maximizing the extent of tumor resection. Since the commonly used deterministic streamline tractography following the principle of tensor deflection underestimates the spatial extent of the white matter tracts, large safety hulls are applied. To optimize the reconstruction of fiber bundle volumes, a new graph-based approach using path finding techniques is presented.

Method: Based on the DTI data set, a weighted graph is reconstructed. Voxels are represented as nodes. Edges are inserted connecting directly neighbored voxels (level 1) or additionally more distant voxels (level 2) allowing a more precise evaluation of principal diffusion direction change between connecting voxels. For each edge, a weighting function is applied that assigns a cost to the respective edge. For reconstruction, a shortest-path-method is used to find connections with minimum cost between defined seed and destination areas. The evaluation is based on 8 T1-weighted and DTI data sets of healthy volunteers (mean age: 27.75 ± 4.82, male/female: 4/4), acquired on a 3T MR-Scanner (Trio, Siemens, Germany) with manual seed region placement outlining the cerebral peduncle and the cortical hand knob area within both hemispheres. The approach has been implemented in C++ within the MeVisLab platform on an Intel Core i7-2600K CPU, 3.4 GHz, 16 GB RAM.

Results: Results of tensor deflection based fiber reconstruction showed solid fiber bundles with a mean bundle volume of 2.104±0.859 cm³ reaching only small parts of the second region of interest outlining the hand knob. Results of the graph-based approach delivered larger bundle reconstructions in all cases with mean bundle volumes of 5.906 ± 0.556 cm³ (level 1) and 7.994 ± 0.981 cm³ (level 2). All obtained reconstructions thereby covered the results of the tensor deflection approach.

Conclusions: A new graph-based shortest-path segmentation was presented showing enlarged reconstructed fiber bundle volumes in contrast to the tensor deflection approach allowing a more precise estimation of their spatial extent.