Investigating the At-Risk Mental State and First Episode Psychosis using Genetic, Cogntive and Multi-modal Neuroimaging Data
: A Multivariate Support Vector Machine Study

Student thesis: Doctoral ThesisDoctor of Philosophy


Numerous studies report significant biological and cognitive alterations in chronic
schizophrenia (ChSz) patients relative to healthy controls (HCs). More recently, similar,albeit less severe, changes have been reported in subjects with a recent first episode of psychosis (FEP), and those at clinical high-risk, referred to as the at-risk mental state (ARMS). The clinical impact of such findings has been limited, however, driven in part by the univariate analyses employed by the majority of studies, which allow inference at the group level only. Support vector machine (SVM) is one alternative multivariate analysis, which, able to provide inference at individual level, has high potential for translation into a clinical setting.
Here, I employed a multimodal approach comprising genetic, structural magnetic
resonance imaging (sMRI), diffusion tensor imaging, functional MRI, and cognitive
data, in order to investigate the capacity of each modality to distinguish FEP and ARMS subjects from HCs, and each other, both at the group, and the single-subject level, using standard univariate and multivariate SVM analyses respectively. Since the clinical potential of SVM is ultimately governed by its classification accuracy, I also performed an empirical comparison of four integrative methods, proposed to enhance classification through data integration.
Collectively, the results provide relative support to the notion that FEP and the ARMS may be characterised by genetic, neuroanatomical, neurofunctional and cognitive alterations similar to those observed in ChSz, albeit less severe. With respect to neuroanatomy, and neurofunction moreover, they suggest such changes may be both subtle, and spatially diffuse. The achievement of only modest classification accuracies, however, suggest that the modalities investigated here have only limited diagnostic power with respect to early-stage psychosis, though it remains that they may be able to provide useful information for predicting conversion to psychosis or treatment outcome, a prospect which could be investigated by future studies.
Date of Award1 Feb 2013
Original languageEnglish
Awarding Institution
  • King's College London
SupervisorAndrea Mechelli (Supervisor) & Paul Allen (Supervisor)

Cite this