Abstract
Purpose :
Dark-adapted humans can detect a handful of photons absorbed in the retina1. Current models of human vision at its sensitivity limit rely on linear retinal processing followed by a nonlinear higher-order decision mechanism. Two recent findings challenge this view: 1) The retinal ON pathway - but not the OFF pathway - integrates sparse signals nonlinearly at visual threshold in primates2, 2) the visually-guided behavior of mice can be directly linked with responses of the ON - but not the OFF - pathway in a photon detection task at the sensitivity limit3. Our aim here was to test if nonlinear signal processing in the retinal ON pathway sets a limit to the detection of weakest flashes in humans.
Methods :
First, we determined an ideal observer's ability to detect dim flashes and to discriminate between two dim flashes of different intensities based on recorded responses of primate ON and OFF parasol retinal ganglion cells (RGCs). Second, we measured the performance of human observers in the detection and in the discrimination tasks in matching conditions with the RGC recordings.
Results :
The performance of human subjects was closely linked with that of the ON - but not the OFF - parasol RGCs in both the detection and the discrimination tasks. For both ON parasol RGCs and humans, the threshold for detecting the dimmest flashes was >1 log unit above the theoretical limit set by physics, whereas their threshold for discriminating between two weak flash intensities was withing a factor of ~2 of the theoretical limit and in agreement with nonlinear retinal signal processing.
Conclusions :
Human psychophysical performance followed closely that of responses of ON but not OFF parasol RGCs at the sensitivity limit. The threshold for detecting the weakest flash is well above the theoretical limit for both humans and for the ON parasol RGCs, whereas the threshold for discriminating between two weak flash intensities is approaching the theoretical limit. Our findings are in agreement with the idea that human vision is optimized for high-fidelity coding of low photon counts at the cost of losing single photons at the sensitivity limit.
1Hecht, S., Shlaer, S., and Pirenne, M.H. (1942). J. Gen. Physiol. 25, 819–840.
2Ala-Laurila, P., and Rieke, F. (2014). Curr Biol 24, 2888-2898.
3Smeds, L, Takeshita, D., Turunen, T., Tiihonen, J., Westö, J., Martyniuk, N., Seppänen, A., and Ala-Laurila, P. (2019). Neuron 104, 576–587.
This is a 2020 ARVO Annual Meeting abstract.