The technique has been applied to different data sets of MR angiography phase contrast and time-of-flight types. For each type two volumes are used to prove the accuracy of the technique. The first type of data is 117 x 256 x 256 (the first two rows of Fig. 9.24) and the second type is 93 x 512 x 512 (the second two rows of Fig. 9.24). First, level sets are initialized by automatic seed initialization. Automatic seed initialization is used in each slice and each slice is divided into windows of size 5 x 5. An average mean is estimated for each class from the average histogram of the volume, and signed distance functions are assigned where each level set function is a collection of Gaussian surfaces added together with a
time step of 0.1 sec. Using this initialization decreases the number of iterations, leading to fast extraction of the vascular tree. The volume segmentation takes about 20 min. on the unix workstation with the super computer. Segmentation results are exposed to the connectivity filter to remove the nonvessel areas. Each volume is visualized to show the vascular tree. The segmentation accuracy was measured to be 94% which is very good for this type of data. The 2D phantom can be modified to be a 3D one simulating the whole volume leading to more accuracy. The results are promising with a good accuracy. This model can be extended to unsupervised case including a parameter estimation capability in future work. Future work will include geometrical features to the segmentation model to enhance the segmentation results.
Was this article helpful?