Wavelet Imaging Features for Classification of First-Episode Schizophrenia

Abstract

Recently, multiple attempts have been made to support computer diagnostics of neuropsychiatric disorders, using neuroimaging data and machine learning methods. This paper deals with the design and implementation of an algorithm for the analysis and classification of magnetic resonance imaging data for the purpose of computer-aided diagnosis of schizophrenia. Features for classification are first extracted using two morphometric methods: voxel-based morphometry (VBM) and deformation-based morphometry (DBM); and then transformed into a wavelet domain by discrete wavelet transform (DWT) with various numbers of decomposition levels. The number of features is reduced by thresholding and subsequent selection by: Fisher’s Discrimination Ratio, Bhattacharyya Distance, and Variances – a metric proposed in the literature recently. Support Vector Machine with a linear kernel is used here as a classifier. The evaluation strategy is based on leave-one-out cross-validation. The highest classification accuracy – 73.08% – was achieved with 1000 features extracted by VBM and DWT at four decomposition levels and selected by Fisher’s Discrimination Ratio and Bhattacharyya distance. In the case of DBM features, the classifier achieved the highest accuracy of 72.12% with 5000 discriminating features, five decomposition levels and the use of Fisher’s Discrimination Ratio.