Decoding the Speed and Motion Direction of Moving Targets Using a Turtle Retinal Patch Model

Cellular level models of neural circuits provide an understanding of the complex neurological phenomena, such as vision. Encoding visual inputs in the visual system of an animal depends on the cellular level properties as well as cell distribution and interconnections in multiple stages. The Retina provides the sensory inputs to the deeper parts of the visual system. Using a model patch of the turtle retina, we show that it is possible to decode the motion direction and speed of a point target moving in the visual space of a turtle, using the spatiotemporal response of a retinal patch. Turtle retina primarily contains two kinds of cells that are functionally different – the intensity sensitive A cells and the motion direction sensitive B cells. The cells are clustered primarily in the vicinity of a visual streak. We analyze the decodability of a circular patch located near the center of the streak. The patch is subjected to a linearly moving point target, as input, that passes through its center along various different angles and speed. The first problem considered is to detect the motion direction of a point target assuming that its speed is known. In the second problem, we assume that both the speed and the motion directions are unknown. The problem considered is to estimate the speed and use this information to detect the motion direction. The two problems are handled using, primarily, the following two methods – one using principal component analysis of the neural signals viewed as a spatio-temporal response of the patch and the other using a suitable pooling process where a selected set of neural responses, over a ‘subpatch,’ are pooled. Counting rate of the pooled process is used for the purpose of decoding under the frame work of inhomogeneous Poisson processes.