Dynamic interactions between sensory and decision-making circuits shape temporal evidence weighting
Perceptual decisions require the brain to integrate sensory evidence over time, yet the temporal weighting of evidence can be flexibly adjusted according to task demands. How such flexibility emerges from interactions between sensory and decision circuits remains unclear. Here, we combined behavioral and neural measurements with recurrent network modeling to investigate the mechanisms underlying temporal weighting of sensory evidence. Macaques performed a motion-discrimination task in which the temporal statistics of sensory evidence favored early, late or uniform integration. Behavioral psychophysical kernels revealed corresponding shifts in the influence of evidence across time, accompanied by systematic changes in motion discriminability and choice-related activity modulation in visual area MT. We developed a recurrent sensory-decision network incorporating reciprocal feedforward and feedback interactions and a temporally structured modulatory input, resembling an urgency signal. The model reproduced the observed behavioral and neural signatures across temporal weighting conditions. Mechanistically, shifting the onset and dynamics of modulatory inputs systematically reconfigured the temporal profile of evidence weighting without substantially altering sensory response dynamics. Choice-related activity in MT can be decomposed in a bottom-up component with the same time-course as the psychophysical kernel, and a top-down component reflecting the decision build-up. These findings suggest that flexible temporal integration can emerge from dynamic interactions between sensory transmission, recurrent decision circuits and temporally structured urgency modulation, providing a mechanistic framework for how neural systems adapt the computation of sensory evidence to changing task demands.