Medical Image Segmentation and Classification Based on Sparse Representation and Dictionary Learnng Algorithms

In this thesis a novel classification approach based on sparse representation framework is proposed. The method finds the minimum Euclidian distance between an input patch (pattern) and atoms (templates) of a learned-base dictionary for different classes to perform the classification task. A mathematical approach is developed to map the sparse representation vector to Euclidian distances. We show that the highest coefficient of the sparse vector is not necessarily a suitable indicator for classification. The proposed algorithm is compared with the conventional Sparse Representation Classification (SRC) framework as well as non-sparse based methods to evaluate its performance. Taking advantage of the introduced classification framework, we then propose a novel fully automated method for the purpose of segmenting different organs in medical images of the human body. Our results demonstrated an acceptable accuracy rate for both classification and the segmentation frameworks. To our knowledge, no other method utilizes sparse representation and dictionary learning techniques in order to segment medical images.