Saccade detection using a particle filter

• Published in: Journal of neuroscience methods Vol. 235; pp. 157 - 168
• Main Authors: Daye, Pierre M., Optican, Lance M.
• Format: Journal Article
• Language: English
• Published: Netherlands Elsevier B.V 30-09-2014
• Subjects: Algorithms; Bayes Theorem; Bayesian estimator; Catch-up saccade; Eye drift; Eye Movement Measurements; Humans; Microsaccades; Nonlinear Dynamics; Pursuit, Smooth; Saccades; Smooth pursuit; Microsaccades; Smooth pursuit; Bayesian estimator; Catch-up saccade; Eye drift
• ISSN: 0165-0270; 1872-678X
• DOI: 10.1016/j.jneumeth.2014.06.020

•Detect saccades using a particle filter as Bayesian estimator.•Remove the baseline velocity signal linked to smooth velocity or eye drifts, allowing a constant threshold for the detection.•Validated with five different paradigms, microsaccades, microsaccades plus saccades with drift, linear target motion, non-linear target motion and free viewing.•Noise sensitivity analysis.
When healthy subjects track a moving target, “catch-up” saccades are triggered to compensate for the non-perfect tracking gain. The evaluation of the pursuit and/or saccade kinematics requires that saccade and pursuit components be separated from the eye movement trace. A similar situation occurs when analyzes eye movements of patients that could contain eye drifts between saccades. This task is especially difficult, because the range of saccadic amplitudes goes from microsaccades (less than 1°) to large exploratory saccades (40°).
In this paper we proposed a new algorithm to detect saccades based on a particle filter. The new method suppresses the baseline velocity component linked to smooth pursuit (or to eye drifts) and thus permits a constant threshold during a trial despite the smooth pursuit behavior. It also accounts for a wide range of saccade amplitudes.
The new method is validated with five different paradigms, microsaccades, microsaccades plus saccades with drift, linear target motion, non-linear target motion and free viewing. The sensitivity of the method to signal noise is analyzed.
Traditional saccade detection algorithms using a velocity (or acceleration or jerk) threshold can be inadequate because of the baseline velocity component linked to the smooth pursuit (especially if the target motion is non-linear, i.e. not constant velocity) or to eye drifts between saccades.
The new method detects saccades in challenging situations involving eye movements between saccades (smooth pursuit and/or eye drifts) and unfiltered recordings.