Classification of complex morphological patterns in Bipolar I Disorder

  • Dr Marcus Zanetti, University of São Paulo, Brazil
  • Dra Maristela Schaufelberger, University of São Paulo, Brazil
  • Dra Deepthi Koka, University of Pennsylvania, United States
  • Dr Luiz Ferreira, Department and Institute of Psychiatry, University of São Paulo, Brazil
  • Prof Paulo Menezes, University of São Paulo, Brazil
  • Dra Marcia Scazufca, University of São Paulo, Brazil
  • Prof Christos Davatzikos, Department of Radiology, University of Pennsylvania, United States
  • Prof Geraldo Busatto, Department and Institute of Psychiatry, University of São Paulo

Objective: Bipolar disorder (BD) has been associated with structural and functional abnormalities of frontolimbic circuits. However, results of structural neuroimaging studies have been highly inconsistent. We aim to evaluate the existence of subtle morphometric abnormalities in bipolar I disorder (BD-I) and to determine the diagnostic power of computer-assisted neuroimaging tools using support vector machine (SVM) classifiers.
Methods: Twenty patients with first-episode psychotic mania and 37 controls were studied with 1.5T structural magnetic resonance imaging (MRI). The images were first registered to a common template through a high-dimensional mass-preserving technique. A multivariate classification method based on SVM was employed to identify the best set of morphological features that discriminate BD-I and control groups. This “morphological signature” was, then, applied at an individual basis using a leave-one-out cross-validation strategy.
Results: The SVM-based classifier ascertained the individuals’ diagnostic status (i.e., whether they were BD-I patients or controls) with an overall accuracy of 87%, 81% of sensitivity and 90% of specificity. The resulting spatial map revealed both gray and white matter abnormalities affecting fronto-limbic-striatal regions in BD-I, mainly in the right hemisphere.
Conclusions: Non-linear, multivariate image analysis of brain MRI is able to detect subtle structural abnormalities even at an early stage of BD-I. The high discriminative power of such methods suggests that the development of auxiliary diagnostic tools is feasible in the near future.