Abstract
The perception of visual motion emerges from a cascade of computations along the magnocellular pathway, from retina through LGN, V1, and MT. Models of motion processing implicitly assume that a smooth and noiseless representation of the visual scene is conveyed by the retina. But the retinal signal is in fact noisy and discretely sampled in space (cells) and time (spikes), and no studies have investigated how efficiently central visual neurons extract information about visual motion from retinal activity. We have previously shown that populations of parasol (magnocellular-projecting) retinal ganglion cells in the macaque retina signal the speed of motion with a precision at least ten times higher than typical measures of behavioral speed sensitivity. This suggests that information is lost along the magnocellular pathway: central readout of the retinal signal fails to exploit all available information from the retina. To identify the origin of this inefficiency, we compared the fidelity of motion signals in neurons in area MT to that in populations of parasol ganglion cells. We used a common set of moving stimuli – drifting bars with Gaussian luminance profiles – to compare the precision of speed estimates obtained from retinal and cortical neurons. We displayed the bars at various speeds while recording from individual neurons or tens of neurons in MT, and computed a maximum posterior estimate of speed from populations of MT cells recorded individually or simultaneously in different experiments. The precision of the speed estimate derived from the MT populations was roughly ten times lower than the precision of speed estimates computed directly from retinal signals. Under the assumption that these recordings reveal the fidelity of the MT population’s speed estimate, the results suggest that the limits on visual motion sensitivity underlying human speed perception arise between the retina and MT.
Acknowledgement: Simons Collaboration on the Global Brain