Brain Image Classification In The Diagnosis Of Nerve Mental Illness

Alzheimer’s disease (AD) is a fatal neurodegenerative disease, which eventually leads to death. The incidence of AD increases with the age, especially for the person elder than65. With the aging the world population, AD not only affects people’s daily activities, but also enlarges the social cost of health care. Based on the rapid development of medical imaging technology, the automatic brain image classification could not only help physicians to make a diagnosis, but also improve the efficiency. So it becomes a hot research topic in the study on Alzheimer’s disease.In this thesis, we focus on the application in the diagnosis of nervous and mental disease via brain image classification, based on three kinds of subjects, which are normal control (CN), mild cognitive impairment (MCI) and Alzheimer’s disease (AD) Firstly, introduce the background and significance of brain image classification on AD and current research worldwide; second, analyze AD symptom, pathology, diagnosis and treatment, also the basic principles of brain imaging, and basic process of brain image classification including feature extraction, dimension reduction or selection and classification; then explore different ways to use multi kernel for brain image classification; finally improve the performance by the two feature selection methods. The contributions of this paper are mainly on:(1) Analyze the relationship between the modalities used and Alzheimer’s disease, and then make two types of research problems:CN vs. AD and CN vs. MCI.(2) In two different types of research problems, explore and compare the different ways.(3) Explore and compare the two feature selection algorithm, select features for different combination of modalities to improve the classification accuracy, and compare the sparsity of the two feature selection methods.The experiment results show that the multi-modalities used are complementary and the combination of them can improve the classification accuracy. The comparison between brain image classification with and without feature selection indicates that using feature selection can further improve the accuracy of classification, and local learning based feature selection outperforms iterative RELIEF algorithm.