An artificial neural network that uses eye-tracking performance to identify patients with schizophrenia

Abstract

Several researchers have underscored the importance of precise characterization of eye-tracking dysfunction (ETD) in patients with schizophrenia. This biological trait appears to be useful in estimating the probability of genetic recombination in an individual, so it may be helpful in linkage studies. This article describes a nonlinear computational model for using ETD to identify schizophrenia. A back-propagation neural network (BPNN) was used to classify schizophrenia patients and normal control subjects on the basis of their eye-tracking performance. Better classification results were obtained with BPNN than with a linear computational model (discriminant analysis): a priori predictions were approximately 80 percent correct. These results suggest, first, that eye- tracking patterns can be useful in distinguishing patients with schizophrenia from a normal comparison group with an accuracy of approximately 80 percent. Second, parallel distributed processing networks are able to detect higher order nonlinear relationships among predictor quantitative measurements of eye-tracking performance.