Using GVF Snake to Segment Liver from CT Images

Abstract

Liver segmentation on computed tomography (CT) images is a challenging task because the images are often corrupted by noise and sampling artifacts. Thus we choose GVF snake to perform the task. Unfortunately, GVF snake use Gaussian function to generate the edge map. We find that this often cause new problems such as blur the liver boundary. To avoid this, a Canny edge detector is a good choice. Another problem during the segmentation is that GVF snake cannot works well with bad initialization, especially when encounter deep concavities. Fortunately we find that if the initial contour can cross the "bottleneck" of the deep concave, it can easily reach the boundary of liver. Thus an algorithm was developed to generate the initial contour automatically. We introduce a new "maximum force angle map" to evaluate the direction variability of the GVF forces. This map can mark up the "bottleneck " and give a trace to run through it. There may be other trace we do not need in the map. With the help of transcendental knowledge about the liver, such as the position, the shape and the Hounsfield unit range of the liver, the correct trace can be found. The contour of this trace is suitable for using as initial contour for GVF snake. By this means we finally segment the liver slice by slice correctly.