Discrimination of auditory hallucination sensitive Schizophrenia patients from resting state functional MRI

Chyzhyk, D. 1 , Graña, M. 1 , Shinn, A. 2 & Ongur, D. 2

1 Computational Intelligence Group, UPV/EHU
2 McLean Hospital, Belmont, Massachusetts; Harvard Medical School, Boston, Massachusetts, US

Abstract The application of machine learning techniques in the field of neurosciences provides to new ways to treat and extract information from neurological data. Application to neuroimage provides new insights and biomarkers that complement those found by statistical inference. This paper deals with resting state funcional Magnetic Resonance Imaging (rs-fMRI) data from Schizophrenia patients with and without a history of auditive hallucinations, and matching controls. We study the binary classification problems arising when considering the discrimination of each pair of subject categories. The steps of the computational process are as follows: (a) rs-fMRI data normalization and preprocessing, (b) reduction of the multivariate data to scalar measures, (c) feature selection, (d) feature extraction, (e) ten fold cross-validation experiments of the classifiers. This study includes the evaluation of four methods to obtain the scalar measures of the rs-fMRI data. Namely we study two functional connectivity measures based on lattice auto-associative memories, respectively. The study includes features from two measures of rs-fMRI spatial coherence and activity. We report very good classification results. We focus on the specific discrimination of Schizophrenia patients with and without auditory hallucinations, showing the brain localizations of discriminant regions for each scalar measure.