Development of Machine Learning-Based Techniques in Psychiatric Neuroimaging

  • Blair Johnston

    Student thesis: Doctoral ThesisDoctor of Medicine


    The primary goal of this thesis is to investigate whether machine learning-
    based methods can be successfully applied to make clinically relevant predictions. These techniques are applied to a range of data such as demographic, socioeconomic and neuropsychiatric variables but primarily to structural and functional magnetic resonance imaging (MRI) data. The main focus of this thesis is to investigate whether the application of these techniques can increase the understanding of psychiatric disorders.

    As MR images contain a large amount of information within each image,
    feature selection techniques, which can identify which brain regions are most
    relevant to the study, are of high importance to maximise the amount of relevant information that is entered into the machine learning approaches. Successfully combining feature selection and machine learning to psychiatric imaging studies has several advantages as the machine learning methods produce output that can separate two or more groups accurately on a subject-by-subject basis or make predictions of a continuous variable and the feature selection provides information on the neurobiology by, for example, highlighting brain regions that are consistently different between groups.

    Two psychiatric disorders are investigated in this thesis: Attention Deficit
    Hyperactivity Disorder (ADHD) and Major Depressive Disorder (MDD). ADHD
    core symptoms include difficulty in sustaining attention, hyperactivity and impulsive behaviour. MDD is a mood disorder that is associated with persistent and disabling symptoms of low mood, anhedonia, hopelessness, guilt, low self-worth, poor concentration, lack of energy, suicidal thoughts and altered appetite and sleep (American Psychiatric Association, 2000). Both disorders do not have any established and reliable diagnostic or prognostic biomarkers so the work undertaken in this thesis aims to identify possible biomarkers using machine learning methods.
    Date of Award2014
    Original languageEnglish
    SupervisorDouglas Steele (Supervisor), Keith Matthews (Supervisor) & David Coghill (Supervisor)

    Cite this