Autoregressive modeling of physiological tremor under microsurgical conditions

Abstract

Tremor was recorded under simulated vitreoretinal microsurgical conditions as subjects attempted to hold an instrument motionless. Several autoregressive models (AR, ARMA, multivariate, and nonlinear) are generated to predict the next value of tremor. It is shown that a sixth order ARMA model predictor can predict a tremor having an amplitude of 96.6 +/- 84.5 microns RMS with an error of 8.2 +/- 5.9 microns RMS, a mean improvement of 47.5% over simple last-value prediction.