Determination of ocular torsion by means of automatic pattern recognition

A new, automatic method for determination of human ocular torsion (OT) was developed based on the tracking of iris patterns in digitized video images. Instead of quantifying OT by means of cross-correlation of circular iris samples, a procedure commonly applied, this new method automatically selects and recovers a set of 36 significant patterns in the iris by the technique of template matching as described by In den Haak et al. Each relocated landmark results in a single estimate of the torsion angle. A robust algorithm estimates OT from this total set of individually determined torsion angles, thereby largely correcting for errors which may arise due to misjudgement of the rotation center. The new method reproduced OT in a prepared set of images of an artificial eye with an accuracy of 0.1 degree. In a sample of 256 images of human eyes, a practical reliability of 0.25 degrees was achieved. To illustrate the method's usefulness, an experiment is described in which ocular torsion was measured during two dynamic conditions of whole-body roll, namely during sinusoidally pendular motion about either an earth horizontal or earth vertical axis (that is "with" and "without" otolith stimulation, respectively).