Review
Using Support Vector Machine to identify imaging biomarkers of neurological and psychiatric disease: A critical review

https://doi.org/10.1016/j.neubiorev.2012.01.004Get rights and content

Abstract

Standard univariate analysis of neuroimaging data has revealed a host of neuroanatomical and functional differences between healthy individuals and patients suffering a wide range of neurological and psychiatric disorders. Significant only at group level however these findings have had limited clinical translation, and recent attention has turned toward alternative forms of analysis, including Support-Vector-Machine (SVM). A type of machine learning, SVM allows categorisation of an individual's previously unseen data into a predefined group using a classification algorithm, developed on a training data set. In recent years, SVM has been successfully applied in the context of disease diagnosis, transition prediction and treatment prognosis, using both structural and functional neuroimaging data. Here we provide a brief overview of the method and review those studies that applied it to the investigation of Alzheimer's disease, schizophrenia, major depression, bipolar disorder, presymptomatic Huntington's disease, Parkinson's disease and autistic spectrum disorder. We conclude by discussing the main theoretical and practical challenges associated with the implementation of this method into the clinic and possible future directions.

Highlights

Support Vector Machine is a multivariate machine learning classification tool. ► Its key strength is that it allows inference at individual rather than group level. ► It has been used for disease diagnosis, transition prediction & treatment prognosis. ► It may lead to computer-based diagnostic and prognostic tools.

Introduction

In the past 20 years the development of Position Emission Tomography (PET) and Magnetic Resonance Imaging (MRI), has allowed the non-invasive investigation of the structure and function of the human brain in health and pathology (Friston, 2009). These techniques have been applied to patients with neurological or psychiatric disorders in order to identify possible biomarkers which could be used for early diagnosis, treatment planning and monitoring of disease progression. This has revealed structural and functional alterations in several disorders including, amongst others, mild cognitive impairment, probable dementia of Alzheimer type, major depression, bipolar disorder, schizophrenia and generalised anxiety disorder (Arnone et al., 2011, Davatzikos and Resnick, 2002, Ellison-Wright and Bullmore, 2010, Etkin and Wager, 2007, Smieskova et al., 2010, Zakzanis et al., 2003). To date however, the results of these studies have had minimal clinical impact and despite much interest in the use of brain scans for diagnostic and prognostic purposes, neurologists and psychiatrists are still forced to rely on traditional and often ineffective diagnostic and prognostic tools. One of the reasons for the limited impact of the findings on clinical practice, is that neuroimaging studies have typically reported differences between patients and controls at group level; in contrast, doctors working in a psychiatric or neurological ward have to make clinical decisions about individuals. For neuroimaging to be useful in a clinical setting therefore, one must be able to make inferences at the level of the individual rather than the group.

Over the past few years, there has been growing interest within the neuroimaging community in the use of analytical methods that allow such inference. One such method is supervised machine learning (ML), an area of artificial intelligence concerned with the development of algorithms and techniques able to automatically extract information from the data (Hastie et al., 2001). Relative to traditional methods of analysis based on the general linear model, the advantages of applying supervised ML to neuroimaging data are twofold. Firstly, supervised ML methods allow characterisation at the level of the individual therefore yielding results with a potentially high level of clinical translation. Secondly, as inherently multivariate approaches, supervised ML methods are sensitive to spatially distributed and subtle effects in the brain that would be otherwise undetectable using traditional univariate methods which focus on gross differences at group level.

Support Vector Machine (SVM) is a specific type of supervised ML method that aims to classify data points by maximising the margin between classes in a high-dimensional space (Pereira et al., 2009, Vapnik, 1995). The optimum algorithm is developed through a “training” phase in which training data are used to develop an algorithm able to discriminate between groups previously defined by the operator (e.g. patients vs. controls), and a “testing” phase in which the algorithm is used to blind-predict the group to which a new observation belongs. The initial applications of SVM to neuroimaging data were primarily aimed at decoding the mental states of healthy volunteers (see Haynes and Rees, 2006 for review); Davatzikos et al. (2005) for example successfully demonstrated that it was possible to discriminate between subjects giving truthful and non-truthful responses with an accuracy of 99.3% based solely on discriminative patterns of brain activity. Since then several studies have used SVM to examine the diagnostic and prognostic potential of neuroimaging in a range of neurological and psychiatric disorders and, to date, a number of promising results have been reported (e.g. Klöppel et al., 2008a, Klöppel et al., 2008b).

In what follows, we begin by providing a brief overview of SVM and its application to neuroimaging data. We then summarise the results of those studies which have examined the diagnostic and prognostic value of structural and functional neuroimaging in neurological and psychiatric disorders before concluding with a discussion regarding the main theoretical and practical challenges associated with actually implementing the use of this methodology into the clinic and possible future directions.

Section snippets

Overview of SVM

Within ML, there are two main approaches that one can take: supervised and unsupervised learning. In supervised learning, one seeks to develop a function which maps two or more sets of observations with two or more, operator defined, categories through an iterative procedure which gradually reduces the difference between the predicted and expected result; subsequently, the algorithm can then be used to assign new, previously unseen, data to one of the predefined categories with a given

A review of SVM studies of neurological and psychiatric disorders

In the past few years, an increasing cohort of studies have used SVM or other pattern recognition methods to investigate possible neuroanatomical biomarkers of neurological and psychiatric disorders. For simplicity, these studies can be divided into three main categories: (i) studies which examine the diagnostic value of neuroimaging data by comparing patients and HCs; (ii) studies which examine the potential of neuroimaging data for predicting the onset of a disease by comparing the brain

Discussion

Neuroimaging studies have revealed structural and functional alterations in several neurological and psychiatric disorders, yet the impact of these findings on clinical practice has been very limited and at the present time there are no objective, biological, markers which can be used to make early diagnosis, prognosis and prediction of clinical outcome in these disorders. SVM is an analytical method which differs from traditional analytical techniques in that (i) it allows the characterisation

Current challenges and future directions

SVM and other supervised ML methods are increasingly being applied to several real world problems, not only in medical research but also in bioinformatics, natural language processing, telecommunications, finance and forensic sciences. Although the application of these methods to neuroimaging data has yielded promising results, there are still significant theoretical and practical challenges for the translational implementation of the results in neurology and psychiatry. Firstly, most studies

Acknowledgements

William Pettersson-Yeo is supported by a PhD studentship from the Medical Research Council (MRC). Andre Marquand gratefully acknowledges support from King's College Annual Fund and the King's College London Centre of Excellence in Medical Engineering. Andrea Mechelli is supported by an Investigator Award from the National Alliance for Research on Schizophrenia and Depression (NARSAD) Independent Award.

References (81)

  • Y. Fan et al.

    Spatial patterns of brain atrophy in MCI patients, identified via high-dimensional pattern classification, predict subsequent cognitive decline

    Neuroimage

    (2008)
  • K. Franke et al.

    Alzheimer's disease neuroimaging initiative estimating the age of healthy subjects from T1-weighted MRI scans using kernel methods: exploring the influence of various parameters

    Neuroimage

    (2010)
  • C.H. Fu et al.

    Pattern classification of sad facial processing: toward the development of neurobiological markers in depression

    Biol. Psychiatr.

    (2008)
  • S. Gauthier et al.

    Mild cognitive impairment

    Lancet

    (2006)
  • Q. Gong et al.

    Prognostic prediction of therapeutic response in depression using high-field MR imaging

    NeuroImage

    (2011)
  • M. Graña et al.

    Computer aided diagnosis system for alzheimer disease using brain diffusion tensor imaging features selected by Pearson's correlation

    Neurosci. Lett.

    (2011)
  • C. Hinrichs et al.

    The ADNI: predicitve markers for AD in a multi-modality framework: an analysis of MCI progression in the ADNI population

    Neuroimage

    (2011)
  • A. Khodayari-Rostamabad et al.

    A pilot study to determine whether machine learning methodologies using pre-treatment electroencephalography can predict the symptomatic response to clozapine therapy

    Clin. Neuropsychol.

    (2010)
  • S. Lemm et al.

    Introduction to machine learning for brain imaging

    Neuroimage

    (2011)
  • A. Marquand et al.

    Quantitative prediction of subjective pain intensity from whole-brain fMRI data using Gaussian processes

    Neuroimage

    (2010)
  • W. Pettersson-Yeo et al.

    Dysconnectivity in schizophrenia: where are we now?

    Neurosci. Biobehav. Rev.

    (2011)
  • F. Pereira et al.

    Machine learning classifiers and fMRI: A tutorial overview

    Neuroimage

    (2009)
  • C. Phillips et al.

    “Relevance vector machine” consciousness classifier applied to cerebral metabolism of vegetative and locked-in patients

    Neuroimage

    (2011)
  • C. Plant et al.

    Automated detection of brain atrophy patterns based on MRI for the prediction of Alzheimer's disease

    Neuroimage

    (2010)
  • H. Shen et al.

    Discriminative analysis of resting-state functional connectivity patterns of schizophrenia using low dimensional embedding of fMRI

    Neuroimage

    (2010)
  • R. Smieskova et al.

    Neuroimaging predictors of transition to psychosis—a systematic review and meta-analysis

    Neurosci. Biobehav. Rev.

    (2010)
  • C.M. Stonnington et al.

    Predicting clinical scores from magnetic resonance scans in Alzheimer disease

    Neuroimage

    (2010)
  • D. Sun et al.

    Elucidating a magnetic resonance Imaging-Based neuroanatomic biomarker for psychosis: classification analysis using probabilistic brain atlas and machine learning algorithms

    Biol. Psychiatr.

    (2009)
  • P. Vemuri et al.

    Alzheimer's disease diagnosis in individual subjects using structural MR image: validation studies

    Neuroimage

    (2008)
  • D. Zhang et al.

    Multimodal classification of Alzheimer's disease and mild cognitive impairment

    Neuroimage

    (2011)
  • D. Arnone et al.

    Magnetic resonance imaging studies in unipolar depression: sdystematic review and meta-regression analyses

    Eur. Neuropsychopharmacol.

    (2011)
  • T.D. Cannon et al.

    The empirical status of the ultra-risk (prodromal) research paradigm

    Schizophr. Bull.

    (2007)
  • G. Chen et al.

    Classification of Alzheimer disease, mild cognitive impairment, and normal cognitive status with large-scale network analysis based on resting-state functional MR imaging

    Radiology

    (2011)
  • C. Chu et al.

    Does feature selection improve classification accuracy? Impact of sample size and feature selection on classification using anatomical magnetic resonance images

    Neuroimage

    (2011)
  • S.G. Costafreda et al.

    Prognostic and diagnostic potential of the structural neuroanatomy of depression

    PLoS One

    (2009)
  • S.G. Costafreda et al.

    Neural correlates of sad faces predict clinical remission to cognitive behavioural therapy in depression

    Neuroreport

    (2009)
  • S.G. Costafreda et al.

    Pattern of neural responses to verbal fluency shows diagnostic specificity for schizophrenia and bipolar disorder

    BMC Psychiatr.

    (2011)
  • Y. Cui et al.

    Automated detection of amnestic mild cognitive impairment in community-dwelling elderly adults: a combined spatial atrophy and white matter alteration approach

    Neuroimage

    (2011)
  • Y. Cui et al.

    s.D.N.Identification of conversion from mild cognitive impairment to Alzheimer's disease using multivariate predictors

    PLoS One

    (2011)
  • C. Davatzikos et al.

    Degenerative age changes in white matter connectivity visualized in vivo using magnetic resonance imaging

    Cereb. Cortex

    (2002)
  • Cited by (807)

    View all citing articles on Scopus
    1

    These two authors have contributed equally to this work.

    View full text