Quantitative evaluation of brain magnetic resonance images using voxel-based morphometry and Bayesian theorem for patients with bipolar disorder

Yong-Sheng Chen, Li Fen Chen*, Ya Ting Chang, Yung Tien Huang, Tong Ping Su, Jen Chuen Hsieh

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

3 Scopus citations

Abstract

The present diagnosis of bipolar disorder (BD), which mainly depends on patients' symptoms and self reports of past history and mood status, may encounter difficulty of distinguishing from unipolar disorder when patients behave depressively in clinic. This work proposes a novel computer-aided evaluation system for bipolar disorder using anatomic magnetic resonance images (MRI) to provide a second opinion for clinician's diagnosis. First we adopt the voxel-based morphometry method to identify brain regions with significant difference between patient and normal control groups as regions of interest (ROIs). Then the MRI data within these ROIs are processed with principal component analysis (PCA) in order to reduce feature dimensionality. Finally, a classification model based on Bayesian theorem together with Parzen-window density estimation in PCA space is constructed to provide the possibility of an individual belonging to the BD patient group. The proposed system reaches 86.8% accuracy in classification. In our experiment, the misdetection rate was zero, and the false alarm was 15.8%. Through appropriate feature analysis and selection method, this computer-aided system can detect the disease of BD and obtain high classification accuracy.

Original languageEnglish
Pages (from-to)127-133
Number of pages7
JournalJournal of Medical and Biological Engineering
Volume28
Issue number3
StatePublished - 1 Sep 2008

Keywords

  • Bayesian model
  • Bipolar disorder
  • Classification
  • Magnetic resonance images
  • Principle component analysis
  • Voxel-based morphometry

Fingerprint Dive into the research topics of 'Quantitative evaluation of brain magnetic resonance images using voxel-based morphometry and Bayesian theorem for patients with bipolar disorder'. Together they form a unique fingerprint.

Cite this