Research On Clinical Application-Oriented Segmentation Algorithms Of MR Images

As the use of chemical substances becomes more and more common and a variety of electronic products are widely used, People’s probabilities of suffering from brain diseases become larger. For this reason, timely diagnosis and regular physical examinations are of great significance. The technology of magnetic resonance imaging (MRI) is widely used with its advantages such as high contrast for soft tissues, high resolution and multiple channels. Faced with a large amount of images that should be processed, doctors have an urgent need for some kind of tools, in order to help complete heavy tasks and improve the efficiency. Because of this, computer aided detections and diagnosis become hot spots in research field.In this thesis, some segmentation algorithms of brain MR images have been researched. The fist chapter introduces the background of this thesis and its significance. Then, an algorithm of skull and background rejection of brain MR images based on threshold segmentation and morphological analysis is achieved. An automatic threshold selection step and a detection of circularity step are added to improve accuracy and practicability. Then some brain tissues segmentation algorithms are analyzed and researched. A robust clustering algorithm is achieved and improved so that it can be robust to both Gaussian noise and pulse noise. A clustering algorithm which has ability to correct the bias field effect is achieved and improved. It can not only get the initial clustering centers automatically but also produce the corrected image and the bias field when it ends. Both the two algorithms can overcome the partial volume effect. Next, an algorithm of brain tumor segmentation based on improved fuzzy connectedness is put forward. It combines clustering methods and the characteristics of MR images that contain tumors and resolves the automatic selection of seeds. What’s more, some fuzzy connectedness algorithms are discussed, the original affinity formula is improved so that it can get better results. Besides, a computer-aided detection platform is designed and established. It can provide some basis for the clinical computer aided diagnosis (CAD) systems aiming for brain diseases in the future. At last, the whole thesis is summarized and possible directions of further study are proposed. The experiments show that the algorithms proposed by this thesis have relatively high accuracy and automatic level.