Brain imaging data are incredibly complex and new information is being learned as approaches to mine these data are developed. In addition to studying the healthy brain, new approaches for using this information to provide information about complex mental illness such as schizophrenia are needed. Functional magnetic resonance imaging (fMRI) and magnetoencephalography (MEG) are two well-known neuroimaging approaches that provide complementary information, both of which provide a huge amount of data that are not easily modelled. Currently, diagnosis of mental disorders is based on a patients self-reported experiences and observed behavior over the longitudinal course of the illness. There is great interest in identifying biologically based marker of illness, rather than relying on symptoms, which are a very indirect manifestation of the illness. The hope is that biological markers will lead to earlier diagnosis and improved treatment as well as reduced costs. Understanding mental disorders is a challenging task due to the complexity of brain structure and function, overlapping features between disorders, small numbers of data sets for training, heterogeneity within disorders, and a very large amount of high dimensional data. This doctoral work proposes machine learning and data mining based algorithms to detect abnormal functional network connectivity patterns of patients with schizophrenia and distinguish them from healthy controls using 1) independent components obtained from task related fMRI data, 2) functional network correlations based on resting-state and a hierarchy of tasks, and 3) functional network correlations in both fMRI and MEG data. The abnormal activation patterns of the functional network correlation of patients are characterized by using a statistical analysis and then used as an input to classification algorithms. The framework presented in this doctoral study is able to achieve good characterization of schizophrenia and provides an initial step towards designing an objective biological marker-based diagnostic test for schizophrenia. The methods we develop can also help us to more fully leverage available imaging technology in order to better understand the mystery of the human brain, the most complex organ in the human body.
Data Mining, Machine Learning, Classification, Neuroimaging, Mental Disorder
Level of Degree
Department of Computer Science
First Committee Member (Chair)
Second Committee Member
Third Committee Member
Cetin, Mustafa Sinan. "New Approaches for Data-mining and Classification of Mental Disorder in Brain Imaging Data." (2015). https://digitalrepository.unm.edu/cs_etds/52