Automatic segmentation of liver using a topology adaptive snake

Most attempts at automatic segmentation of liver from computerised tomography images to date have relied on low-level segmentation techniques, such as thresholding and mathematical morphology, to obtain the basic liver structure. The derived boundary can then be smoothed or refined using more advanced methods. In this paper we present a method by which a topology adaptive active contour model, or snake, accurately segments liver tissue from CT images. The use of conventional snakes for liver segmentation is difficult due to the presence of other organs closely surrounding the liver. Our technique avoids this problem by adding an inflationary force to the basic snake equation, and initialising the snake inside the liver. Once the user has initialised the snake for one CT slice, the starting locations for other slices in a dataset are determined automatically from the center of gravity of the segmented area of previous slice. We present results from over 500 images, covering 4 different healthy datasets, and each liver slice is segmented in 2D before being compared to the equivalent segmentation performed by hand. Statistical analysis of the datasets shows that, in each case, there is no significant difference between the areas and the snake-segmented liver to the areas of hand segmented liver, here treated as the gold standard.