Discriminating unipolar and bipolar depression by means of fMRI and pattern classification: a pilot study

  • Dominik Grotegerd
  • Thomas Suslow
  • Jochen Bauer
  • Patricia Ohrmann
  • Volker Arolt
  • Anja Stuhrmann
  • Walter Heindel
  • Harald Kugel
  • Udo Dannlowski
Original Paper

DOI: 10.1007/s00406-012-0329-4

Cite this article as:
Grotegerd, D., Suslow, T., Bauer, J. et al. Eur Arch Psychiatry Clin Neurosci (2013) 263: 119. doi:10.1007/s00406-012-0329-4

Abstract

Bipolar disorders rank among the most debilitating psychiatric diseases. Bipolar depression is often misdiagnosed as unipolar depression, leading to suboptimal therapy and poor outcomes. Discriminating unipolar and bipolar depression at earlier stages of illness could therefore help to facilitate efficient and specific treatment. In the present study, the neurobiological underpinnings of emotion processing were investigated in a sample of unipolar and bipolar depressed patients matched for age, gender, and depression severity by means of fMRI. A pattern-classification approach was employed to discriminate the two samples. The pattern classification yielded up to 90 % accuracy rates discriminating the two groups. According to the feature weights of the multivariate maps, medial prefrontal and orbitofrontal regions contributed to classifications specific to unipolar depression, whereas stronger feature weights in dorsolateral prefrontal areas contribute to classifications as bipolar. Strong feature weights were observed in the amygdala for the negative faces condition, which were specific to unipolar depression, whereas higher amygdala features weights during the positive faces condition were observed, specific to bipolar subjects. Standard univariate fMRI analysis supports an interpretation, where this might be related to a higher responsiveness, by yielding a significant emotion × group interaction within the bilateral amygdala. We conclude that pattern-classification techniques could be a promising tool to classify acutely depressed subjects as unipolar or bipolar. However, since the present approach deals with small sample sizes, it should be considered as a proof-of-concept study. Hence, results have to be confirmed in larger samples preferably of unmedicated subjects.

Keywords

Depression Bipolar disorder fMRI Amygdala Machine learning Pattern classification 

Supplementary material

406_2012_329_MOESM1_ESM.pdf (364 kb)
Supplementary material 1 (PDF 363 kb)

Copyright information

© Springer-Verlag 2012

Authors and Affiliations

  • Dominik Grotegerd
    • 1
  • Thomas Suslow
    • 1
    • 2
  • Jochen Bauer
    • 1
  • Patricia Ohrmann
    • 1
  • Volker Arolt
    • 1
  • Anja Stuhrmann
    • 1
  • Walter Heindel
    • 3
  • Harald Kugel
    • 3
  • Udo Dannlowski
    • 1
    • 4
  1. 1.Department of PsychiatryUniversity of MünsterMünsterGermany
  2. 2.Department of Psychosomatic Medicine and PsychotherapyUniversity of LeipzigLeipzigGermany
  3. 3.Department of Clinical RadiologyUniversity of MünsterMünsterGermany
  4. 4.Department of PsychiatryUniversity of MarburgMarburgGermany