A cortico-collicular model of saccadic target selection

Abstract

Human saccadic programming is subject to bottom-up and top-down influences. A neural network model was developed to show how the two processes may interact to produce eye movements. The model consisted of modules loosely corresponding to presaccadic processing in the superior colliculus (SC) and cortex. The collicular module provided a fast and unprocessed interpretation of the visual scene while the cortical module proceessed inputs based on task dependent variables such as target/nontarget differentiation. Topographic connections between units in the cortical and collicular modules allowed target information to selectively enhance locations within the SC. Winner-take-all competition between active SC populations was then used to suggest a specific oculomotor target. The performance of the model was tested against human oculomotor data. The results indicated a qualitative similarity between human and simulated fixations. In particular, simulated saccades to double target stimuli exhibited realistic averaging behavior. A pronounced interaction between saccadic accuracy and latency was also observed. Simulated eye movements became more accurate with longer saccadic latencies. The correspondence between actual and simulated behavior suggests that a top-down component may affect saccadic programming by selectively enhancing target locations within a bottom-up computation.