This dissertation addresses the segmentation and classification problem of normal and abnormal structures in the human body. Due to the boundary ambiguity between regions in medical images, organ segmentation is a challenging task, and it requires prior knowledge for accurate segmentation. The segmentation objectives in this dissertation are to develop fully automatic methods for anatomical organ segmentation using prior knowledge. Prior knowledge is incorporated in terms of local and global image features. Two novel strategies are proposed. The first one is based on global image features. The second strategy is combining the local and global image features using both the level set and the dictionary learning methods.
