Figure 18.

Comparison of performance of mouse ganglion cell and model. The experimental task was to detect a dim flash of light at different backgrounds, measured as the threshold photon flux (R*) that produced a response equal to the standard deviation (Dunn et al., 2006). Left, model of rod convergence to the ganglion cell, including 4096 rods, each converging through a synaptic nonlinearity, summation, and gain control. Right, gray trace shows threshold of ganglion cell (+/−SEM), and black trace shows threshold photon flux of model. At low backgrounds, threshold of the ganglion cell was ~3-fold higher than the model. At higher backgrounds, as gain was reduced, performance of the model approached the performance of the ganglion cell. This suggested that an extra noise source must exist in the pathway, for example, at the rod bipolar to AII amacrine synapse. Threshold curves were identical for a site after summation (1) and for a site after the gain control (2). The reason is that the gain control was noiseless. Performance of the array of rods was greater than the ganglion cell. The ideal observer allows an unbiased comparison between real cell and model at different stages. Compare this model to Figure 17. Redrawn from Dunn et al. (2006).