gms | German Medical Science

Objective: We used training experiments with simulated retina implant (RI) perception to investigate how object recognition can be improved for RI-patients. We first investigated the efficiency of different retrieval schedules. In two more experiments, we investigated training transfer across tasks.

Methods: Normal-sighted participants (N=16 per experiment) took part in three experiments. RI-perception was simulated with the Pulse-to-Percept software (Beyeler et al., 2017). Grey-scale photographs from eight categories of common objects were transformed, using a resolution of 1369 stimulating electrodes and a visual field of 8° visual angle and minimal axonal stimulation (lambda = 0.1) to simulate subretinal stimulation. Recognition test trials were spaced at equal or expanding intervals. The task was either to select the corresponding verbal label for a simulated object image (labeling task) or a verbal object label was shown and the fitting object picture had to be selected (reversed labeling task).

Results: Object recognition training led to significant improvement of object recognition in both tasks that lasted one week in the labeling task, but not the reverse labeling task. Transfer was observed from the labeling to the reverse labeling task but not v.v. Equal and expanding retrieval intervals led to comparable recognition at follow-up.

Conclusions: Object recognition training with simulated RI perception improved recognition considerably. Object learning occurred over different views and exemplars and was stable at one-week follow up. Learning transferred from a labeling to a reverse labeling task, but not v.v. Training regimes for patients should build on these findings, particularly with regard to task selection. RI-patient training data are needed to validate the simulation data.