Seed points, i.e., starting points for fiber tracking, are defined manually or based on an atlas. In the manual configuration of seed points, ROI is first specified, and then seed points are automatically placed with in the ROI.

Fig. 3.22 Basic scheme for DTT

(2) . Diffusion tensor calculation

For each seed point, a diffusion tensor is calculated with DWI signal values including diagonalization for obtaining eigenvalues and eigenvectors. The DTT approaches assume that fiber orientation corresponds to the orientation of maximum diffusion coefficient obtained by the eigenvector of the maximum eigenvalues.

(3) . Short step movement

Move in a short step distance along the direction of the maximum eigenvalue of the diffusion tensor. Note that there are two directions for initial movement at the seed point, that is, two trajectories per seed point are yielded. After the initial movement in two ways, only one direction is selected so that the trajectory is smoother.

(4) . Iteration of tracking

Procedures (2) and (3) are repeated iteratively until tracking termination conditions are satisfied. These conditions are low diffusion tensor anisotropy, large angle of direction change, etc.

As described earlier in this chapter, the DTT based on the single tensor model is inadequate in situations involving fiber crossing. Therefore, the ODF-based tractography for DSI or QBI data is approaches are preferable as they have more options to determine tracking directions based on the local maxima of ODF. This is a natural extension of the DTT in the local approach. The local approaches of DTT and ODF-based tractography are subcategorized as “deterministic approaches,” to distinguish them from another branch of local approaches known as “probabilistic approaches” [35]. In deterministic approaches, tracking is performed only once per seed point because the trajectory is identical if the tracking scheme and termination

Fig. 3.23 Probabilistic tractography example

conditions are identical. In contrast, probabilistic approaches perform tracking several times from a seed point. This is because the tracking direction changes slightly each time a random sampling in the diffusion tensor or ODF is used as the probability function. Consequently, higher sensitivity for minor fiber structures occurs, resulting in false-positive structures. An example is shown in Fig. 3.23 and is based on the notion of uncertainty of fiber direction determination [71].

The global approaches, including model-based approaches using a fiber tract atlas, are quite different from local approaches. That is, in those approaches, smoothness of trajectory or similarity to the prior trajectory is considered. Figure 3.24 shows an example of global approach proposed by Reisert [237].