Bio-inspired methods for analysis and classification of reading eye movements of dyslexic children

The paper describes real practical problem of analyzing diagnostic significance of dyslexic eye movements. The biologically inspired methods were used and compared with classical methods of artificial intelligence. Eye movements of 52 school children were measured using videooculographic (VOG) technique, during a reading task. There were three groups of subjects – normal readers, retarded readers and dyslexics. The main goal was to analyze the possibility of dyslexia detection only from the eye movement signal. Time and frequency domain features were extracted and subset of significant features was chosen by a simple feature selection method. The selected feature subset was visualized using a self-organizing map (SOM). Clusters were formed by the SOM proving that proposed methodology is suitable for automatic dyslexia detection. Next, the artificial neural network classifier was used for supervised learning and the performance was compared to the results obtained for nearest neighbour classifier or the Bayes classifier.