Abstract
Most visual information is acquired during fixation. While saccades are known to be driven both by sensory information and cognitive factors, the role of cognitive factors in fixational eye movements (FEM) is largely unknown. To probe these influences, subjects were asked to discriminate between two foveally-presented letters (1.5 degrees) superimposed on visual noise whose contrast was adjusted to reach ~75% correct performance. Eye movements were monitored by a digital dual-Purkinje-image eyetracker. Trials were organized in blocks with a fixed letter pair, each letter presented in 40% of the trials. 20% of trials contained only noise. We built a computational model to choose letter pairs for which FEM trajectories were likely to affect performance. The model used previously-measured FEM trajectories to generate time-varying firing rates of LGN neurons from linear spatiotemporal receptive fields. For each trajectory, a Bayesian decision stage pooled across neurons. This yielded a prediction of discrimination performance for each candidate letter pair and each FEM trajectory, which was similar, on average, to human performance. Among multiple letter pairs examined, EF and HN discriminations showed the largest dependence on the FEM trajectories. Data from two subjects suggests several influences of cognitive factors on FEMs. For microsaccades, starting and landing points differed: both tended to be below center in EF trials, but near center for HN. Microsaccade directions were more often upward for EF, but more often diagonal for HN. Drift statistics also differed: the velocity distribution was biased towards vertical for EF, but more isotropic for HN. Interestingly, most of these findings were present in the noise-only trials of each block, suggesting that they can be driven by task knowledge, independent of the sensory stimulus.