The Statistical Detection of Threshold Signals in the Retina

1. Photographic records of impulses from single ganglion cells in the cat's retina were made while the retina was stimulated by flashes occurring once a second. Ten flashes at each of several intensities near threshold were used. 2. For the purpose of statistical analysis, the number of impulses (x) falling within a critical period following each flash was used as an index of the response. Histograms of x were plotted and used to calculate rates of transfer of information by the ganglion cell for the case of an ideal experiment, the yes-no choice, in which flashes of intensity I and blanks are to be distinguished. 3. The information rate increased (a) with increasing stimulus intensity and (b) with the number of identical flashes or blanks presented successively in a block. The intensity chosen as threshold by the experimenter, who observed the impulses visually and aurally, corresponded to an average information rate for single flashes of 0.7 bit/flash, compared to the maximum possible rate of 1 bit/flash. A threshold intensity giving 0.4 or more bit/flash, if presented in blocks of six identical flashes, corresponded to 0.95 or more bit/block, or near certainty. Thus the calculation of information rates using the index x provides an estimate of threshold at least as sensitive as those obtained during an experiment, which were made only after observing the responses to five to ten flashes of the same intensity. 4. The index x has statistical properties similar to those of the "index of neural activity" used by Tanner and Swets (1954) in their statistical model of human vision, and represents a possible physical interpretation of their index. However, x gave values (0.5 to 1.5) of the parameter called the slope which were consistently smaller than their values (2.1 to 3.1).