Cocaine Dependent Classification on MRI Data Extracting Features from Voxel Based Morphometry

In this paper, we present a method to discriminate cocaine dependent patients and healthy subjects using features computed from structural magnetic resonance imaging (MRI). After image preprocessing, we compute voxel based morphometry (VBM) applying Gaussian smoothing with three different full width at half maximum (FWHM) kernel sizes. VBM clusters guide the feature extraction process used to classify subjects as cocaine dependent patients or healthy controls. We apply five well known classifiers from the WEKA platform. Classification results are good reaching accuracy, sensitivity and specificity values above 90%. It is possible to apreciate that as the smoothing kernel size grows, the features are less discriminative, but the VBM clusters identifying differences between both groups are bigger. We also obtain the location in the brain of the features selected and compare them with findings in the literature.