Research On The Segmentation Of Prostate Ultrasonic Images

With the progress of medical image processing technology and the development of ultrasonic medical, more and more prostate diseases were found and this arouse extensive attention. Comparing Ultrasonic Image (US) with CT and MRI, we can find the quality of the Ultrasonic Image is very bad for plenty of noise. In order to help doctor to check the tissue of prostate and diagnose illness exactly, it is increasingly demand that how to realize the automatic segmentation from ultrasonic image.The research object in this dissertation is prostate ultrasonic images. Firstly, the principle of ultrasound devices and the B ultrasonic imaging have been introduced, and then the anatomic structure of prostate and its pathology have been studied. The particularities of prostate ultrasonic image have been researched in detail as well.In the pretreatment of these images, the Speckle noise model has been improved in this dissertation. On the basis of traditional methods of neighborhood averaging and low-pass filtering, the methods of Adaptive Histogram Enhancement and the improved adaptive Weighted Median Filter have been presented, and the noise has been reduced greatly.After the study of the traditional algorithms of feature extraction and boundary tracking, the edge detection method based on combination of γ-correction and Sobel edge operator has been realized. After the discussion of several kinds of optimum threshold segmentation methods, a multi-feature vector space and three new criteria (Global Comparison Detection, Geography Priority Privilege, Equal Opportunity for Competence) are developed for region growing control, a new region growing method is brought forward. At last the region splitting and merging, region clustering, neural networks, SNAKE active contour model et al have been discussed.Both the 256-color ultrasonic image processing system and 24-bits-color ultrasonic image processing system has been designed based on the methods above. The results of computer simulation show that the methods presented in this dissertation have the good effects on the segmentation of ultrasonic image.