Segmentation Method For Medical Image Based On Improved Grabcut

Segmentation of medical images has a lot of interferences because of the low contrast and fuzzy boundaries. It's hard to get perfect effect using present image segmentation methods, so we put forward an improved algorithm based on GrabCut and Gaussian mixture model (GMM) in this paper in order to obtain simplify interactive operation and better segmentation precision. We extend the GrabCut approach in 2 respects. Firstly, the initial GMMs of foreground and background were obtained by training sets, which could improve the algorithm's convergence rate. Secondly, the segmentation was restricted by the figure of foreground from training. Experimental results showed that compared with the traditional GrabCut algorithm, our proposed algorithm can simplify interactive operation (t=14.33, P<.01) and improve the segmentation speed (t=16.77, P<.01). In addition, in respect of segmentation precision, our proposed algorithm was obviously better than the traditional algorithms such as Graph Cut, GrabCut and Lazy Snapping. (F=149.546, P<.01). The improved algorithm we proposed in this manuscript is especially suitable for processing large-scale medical images.