Study On Medical Image Processing In Prostate Puncture Guidance

Prostate cancer constitutes the most common malignant tumor in men’s genital system. As the International Agency for Research on Cancer reported, prostate cancer is the fifth most common cancer in the world and the second most common lethal cancer in men. The transrectal biopsy is the most reliable diagnosis technique to detect prostate cancer, while ultrasonic imaging and magnetic resonance imaging are assistants for clinic diagnosing, monitoring and guiding. It is useful that prostate images from different modes are effectively processed by computers, which supports clinic diagnosis, treatments and researches. For instance, it can be employed to enhance the function of a computer aided diagnosis system, guide the prostate biopsy procedure and so on. Therefore, the objective of the dissertation is to find suitable methods for prostate image processing, in order to provide valid guidance for prostate biopsy. The dissertation contains detecting prostate boundaries in ultrasonic images and magnetic resonance images, and registering these two modes of prostate images.The first section of the dissertation is prostate boundaries extraction from transrectal ultrasound images by a discrete contour point set method based on multi-feature and multi-level selection. First, a series of boundary points are picked out during the multi-feature and multi-level selection procedure. The points in the point set form a discrete contour to local the prostate. After that, a curve evolution model is applied to refine the discrete contour. For each image, the method automatically obtains a specialized discrete contour, which accommodates the image more than the statistical contour. So the following adjustment needs less calculation, and the absence of the statistical contour means no data processing for plenty of training images.In the second section, the dissertation focuses on a method based on the active shape with adaptive texture distribution, in order to detect prostate contours in magnetic resonance images. First, a prostate region of interest is determined through image classification and fitting. At the same time, shape parameters are estimated for the active shape. Then the texture coincidence measure is built with multi-features, while the active shape is divided into two portions, the texture distribution shape and the supplementary shape. The two components make up a new active shape to make it more capable of searching and matching during the segmentation. Finally, the segmentation of a prostate image starts from an optimized initial estimation of the active shape, and the searching procedure is adjusted according to the texture distribution shape and the supplementary shape. The method is an effective modification of the active shape, and it can automatically extract prostate boundaries from magnetic resonance images with high enough accuracy.The registration of the prostate ultrasonic image and the magnetic resonance image is the third part of the dissertation, which is based on3D prostate model analysis and contour point matching. A3D prostate model is built from slices of the magnetic resonance images, and its finite elementary analysis yields a deformable information model. The slice to be registration is semi-automatically located with the angle information from a specialist. Then the magnetic resonance slice is registered to the ultrasonic slice by a point match method with the deformable constrain of the3D prostate model. During the procedure, it costs longer time to construct and analyze the3D prostate model, while the registration is faster. The condition conforms to the need of prostate biopsy guidance, and the procedure is supposed to be carried out as follows. During preoperative examination, magnetic resonance images are scanned to extract the boundaries and build the prostate model, and the results of prostate contours and analysis are saved. When intraoperative biopsy comes, a prostate contour is detected in the transrectal ultrasound image, and the magnetic resonance slice is registered to provide better image guidance.