Nonlinear neural circuit model accounts for nonhuman primates’ choice behaviour and LIP neuronal activity in perceptual decisions uncoupled from motor actions

Perceptual decision-making has often been studied using sensory-to-motor transformation tasks. Shushruth et al. (2022) showed in a visual motion discrimination task (Fig. A) that neurons in the lateral intraparietal cortex (LIP) of two monkeys exhibited delayed evidence-based responses, activating only upon choice target onset after the motion stimulus (Fig. B-C). The LIP neurons were hypothesised to have retrieved information via short-term memory. However, how this was achieved remains unclear.In this work, we extended the nonlinear mean-field model of Wong and Wang (2006) by incorporating a hypothesised unobserved upstream neural layer that encoded left/right motion direction and accumulated evidence over time (Fig. D). Its outputs were fed into the model’s LIP layer, which encoded the selection of the choice targets (Fig. D). The model readily recapitulated the choice accuracy and (correct and error) reaction time (Fig. B), and the LIP neuronal activity profiles of the monkey’s data (Fig. E). In particular, the LIP layer showed evidence-based activation only upon choice target onset, which acted as a gate: the sensory-evidence accumulating layer was able to maintain the choice made in the absence of the stimulus while influencing the LIP layer (Fig. D). Our results offer a dynamical neural circuit perspective on how sensory information is integrated, cached and acted upon during perceptual decision-making uncoupled from motor action.