Hidden Markov models for analysis of eye movements of dyslexic children

The paper describes an application of hidden Markov models to dyslexia detection from eye movements. Eye movements of reading-age dyslexic and control children are measured, pre-processed and hidden Markov model with two hidden states is trained on velocity time series for each child. The two states of the model correspond to two component of the eye movements signal - fixations and saccades. The elements of transition matrix are further used one by one as features for 1-dimensional linear Bayes classifier. It is shown that this method applied to eye movements during the simplest non-verbal task can lead to relatively high performance. Thus, we propose this feature extraction for a more sophisticated systems which would be able to detect dyslexia in pre-school children.