gms | German Medical Science

Objective: To describe the patterns of visual exploration of letter-like images during a simulation of artificial vision.

Materials and Methods: We trained three non-human primates to recognize letters of the alphabet in a two-alternative forced-choice task where the cue letter was presented through a simulation of artificial vision. Letters of the Stelio font (a generalization of the Sloan set) were presented at various sizes, and through simulations of four prosthesis designs with varying electrode counts. Human subjects also performed this task, but with letter-like glyphs created to have similar visual statistics to the 26 letters in the Stelio font.

Results: The three non-human primates (NHP) tended to preferentially examine vertexes and intersections, using a stereotypical pattern that differed from letter to letter, but was conserved across phosphene patterns. They exhibited a substantial training effect whereby fewer and fewer saccades were performed over time as they learned the task. Humans had a similar pattern of exploration, but a much more rapid training period. The latency to first saccade during the cue period was anti-correlated with the number of phosphenes in the patterns (longer latencies were associated with patterns with fewer phosphenes) for both groups of subjects.

Discussion: Each NHP developed an individual strategy for letter recognition, exhibiting gaze patterns that were focused on characteristic features of each letter based on both verticies and stroke features. The anti-correlation between latency of first saccade and phosphene density is consistent with our previously hypothesized information diffusion model that builds illusory contours between phosphenes.

Acknowledgment: This work was supported by the NIH under EY019679, EY027888, and the William M. Wood Foundation, Bank of America, Trustee.