| Medical image is the key of qualitative and quantitative analysis for hum antissue by doctor. The task of medical image segmentation is separating the focus andnormal area. The result is accurate or not which directly affect doctor’s diagnosis andthe subsequent processing of the images. Medical image, however, usually has astrong noise, weak contrast and vague organizational boundaries, that bring a greatdeal of difficulties to segmentation task.Because of the plasticity of energy function and its global optimality, GraphCuts has developed rapidly as a new image segmentation method in graph theory; andpriori information can be added to it by human interactions, with which, it couldguide the image segmentation well. Therefore, it has better applicability in complexmedical image segmentation. In medical image research, this paper made someimprovements by adding SUSAN edge character, on the basis of Graph Cutsalgorithm, in order to achieve a more accurate segmentation results. The main jobswe complete are as follows:1. Unlike the past differential edge detection operators, SUSAN operator do notuse differential operation as the means of edge extraction. It has a better effect ofextracting edge by introducing more local information, from medical image whichhas noise and weak boundaries. This paper improves the stability, adaptivity and noiseimmunity of the SUSAN operator based on original one, in order to make it moresuitable for extracting edge from medical image.2. So as to introduce the new SUSAN character to replace the gray value, weoptimize and improved the energy function of Graph Cuts. It makes the newalgorithm more suitable for medical image segmentation, and achieves bettersegmentation results. Through experiments, we found that the new algorithm cansegment the interested organizations from the prostate MRI and other medicalimages more accurately than original Graph Cuts method did. It provides a goodbasis for subsequent image processing and clinical diagnosis of the doctor. |
PDF Full Text Request