Predictive classification of individual magnetic resonance imaging scans from children and adolescents

Neuroimaging techniques are increasingly being explored as potential tools for clinical prediction in psychiatry. There are a wide range of approaches which can be applied to make individual predictions for various aspects of disorders such as diagnostic status, symptom severity scores, identification of patients at risk of developing disorders and estimation of the likelihood of response to treatment. This selective review highlights a popular group of pattern recognition techniques, support vector machines (SVMs) for use with structural magnetic resonance imaging scans. First, however, we outline various practical issues, limitations and techniques which need to be considered before SVM's can be applied. We begin with a discussion on the practicalities of scanning children and adolescent participants and the importance of acquiring high quality images. Scan processing required for inter-subject comparisons is then discussed. We then briefly discuss feature selection and other considerations when applying pattern recognition techniques. Finally, SVMs are described and various studies highlighted to indicate the potential of these techniques for child and adolescent psychiatric research.