Abstract
One significant limitation of electronic visual prostheses is that they cause simultaneous, rather than complementary firing within on- and off-center retinal cells. Here, using ‘virtual patients’ – sighted individuals viewing distorted input – we examine whether plasticity might compensate for the resulting neuronal population coding distortions. Six participants were trained on a video game using real-time dichoptic presentation of original and contrast-reversed stimuli as a proxy for inappropriate on- and off-cell stimulation. Each greyscale image frame was filtered in Fourier space such that regions producing on-responses in one eye produce off-responses in the other eye. The goal of the video game (based on ‘Fruit NinjaTM’) was to correctly ‘swipe’ target objects, while avoiding distractor objects. Participants played for a total of 30 one-hour sessions. An object discrimination task containing novel objects was interleaved every 5 sessions to measure perceptual learning. A separate control group of subjects was given only the object discrimination task. All subjects in the gaming group continuously improved across each object discrimination task, with an average impovement in dprime scores of 1.97 (ranging between 1.65 – 2.21) over 5 sessions. Participants in the control group (n = 3) showed slightly less improvement (m = 1.6, ranging between 1.40 – 1.80). We used four transfer of learning tests to examine the mechanisms of learning: (1) monocular presentation, (2) binocular presentation but the target was only in one of the two eyes, (3) trained filters were swapped across each eye, (4) 1/f noise replaced the contrast reversed image information. Results suggest that performance improvements were due to participants learning to extract relevant from irrelevant image information, in a non-eye-selective manner. This work suggests that it is possible to learn to learn novel decoding of unnatural visual cell population responses in adulthood.