Brain Segmentation Method

Segmentation is achieved in three stages as shown in Fig. 2: removal of the background using intensity histograms, generation of an initial mask that determines the intracranial boundary with a nonlinear anisotropic diffusion filter, and final segmentation with an active contour model [22]. The use of a visual programming environment such as WiT [3] makes prototype development more convenient by allowing some exploration [5]. Preferably the T2-weighted image is used; otherwise the PD-weighted or T1-weighted image may also be used for segmentation. RF inhomogeneities are addressed by the smoothing obtained with the nonlinear anisotropic diffusion, which also reduces the intensity of regions that do not belong to the brain.1 In the third stage, the relative insensitivity of the active contours to partial volume effects provides consistent edge tracking for the final segmentation. This sequence of operations provides a relatively robust approach that results in good segmentation even in the presence of RF inhomogeneity, where simple thresholding techniques would be difficult to use.

Two types of prior knowledge are needed in the second stage, wherease the first and third stages do not require prior information. The first type of prior information relates to tissues other than the brain, for example, the scalp and eyes. Knowledge of the anatomic location of the different structures indicates that the centroid of the brain has to be close to the centroid of the entire slice image. The fact that the brain has a

1Sled's method can be used to correct severe inhomogeneities [28].

FIGURE 2 A simplified data flow diagram representing automatic intracranial boundary detection. Reprinted from M. S. Atkins and B. Mackiewich. Fully automatic segmentation of the brain in MRI. IEEE Transactions on Medical Imaging, 17(1):98-107, February, 1998. © 1998 IEEE.

Was this article helpful?

0 0

Post a comment