Machine learning on neuroimaging of brain disorders

The diagnosis of Schizophrenia and related psychoses is based on interview and clinical symptoms. The diagnosis could be challenged by the complex and heterogeneous symptoms as well as many confounders such as sex and age of the patients. Diffusion Tensor Images(DTI) are important brain medical data which can be the evidence for the diagnosis of schizophrenia. The dissertation obtains the necessary features by pre-processing the DTI. The machine learning methods are adopted to select features which can discriminate patients with Schizophrenia from healthy people. The classifiers are trained based on four machine learning algorithms and applied on the testing sets to evaluate the availabilities and accuracies of these algorithms. The dissertation calculates the testing accuracies of four algorithms based on different selected parameters and different types of samples. The dissertation utilizes the linear correlation among same types of images to select features. The dissertation proposes that Probability Neural Network can reach the highest accuracy based on a specific type of parameters. The discrepancy between genders shows that gender can influence the diagnosis of Schizophrenia.Master of Science (Signal Processing