Application of artificial neural networks (ANNs) to complex oscillations in the human pupil light reflex

Abstract

Experimental, yet strongly aperiodic, data have been obtained from the human pupil light reflex. The aperiodic behavior in this reflex is similar to that in chronic granulocytic leukemia victims; data on the latter, however, are not readily available. A physical model of this reflex has been developed, but is too regular to predict aperiodicity in pupil area oscillations accurately. Artificial neural networks (ANNs) have been trained to do time-series predictions of this aperiodicity with greater accuracy. It is shown that the physical model should be used in setting the input neurons of the ANN, that the physical model and the ANN complement each other, and that ANNs have a role in identification of dynamical diseases in particular and aperiodic (if not chaotic) behavior in general.<>