GMM Based Detection of Schizophrenia Using Eye Tracking

We present a method for schizophrenia disorders detection that utilize Rorschach Inkblot Test in combination with an eye-tracker. Both the trajectory an image is scanned through as well as an overall focus that is paid to defined regions of the image are considered. Using well know Markov chain modelling these static as well as dynamic features can be easily expressed at the same time in a compact way from unequal length observations. In the classification stage a generative model base method (GMM) was deployed and tested with several strategies and settings. All experiments were accomplished on dataset consisting of 44 individuals (22 healthy and 22 diagnosed patients). On average the achieved accuracies were mainly in the interval from 70% to approx. 80% and the best observed score was 79.2%. Combinations of features, strategies and settings of classification techniques significantly influence the final outcome. However on average mean adapted GMM models scored best either using transition matrices or vectors of final probabilities in case of diagonal covariance matrices.