8 122 8. General discussion Functional prototyping of software and hardware choices in a naturalistic mobility task revealed a nuanced perspective on the optimal number of electrodes and the ideal scene processing software. The results showed an interaction effect between electrode count and the scene simplification, suggesting that the functional outcomes depend on a multitude of parameters. An exploration of automated evaluation strategies revealed that efficient phosphene encodings can be learned using deep-learning-based parameter optimization and virtual implant users. The results indicate that phosphene encodings can be constrained with realistic restrictions such as sparse electrical stimulation. Furthermore, the results show that the encoding can be guided towards task-relevant information and account for specific implant characteristics such as arbitrary electrode locations. The proof-of-principle results of our end-to-end simulations underscore the potential of deep-learning for the optimization of the multitude of contextual parameters in prosthetic vision. The development of simulation models with improved biological plausibility was found to better reflect empirical findings reported in the prior brain stimulation literature. The simulation of specific phosphene characteristics improves the translational validity and helps to address clinical questions (e.g., regarding the electrical stimulation parameters) with more explicit solutions. Experimental results with a more realistic simulator revealed significant benefits of including an eye-tracking system when possible. The studies in this dissertation underscore the importance of including biological models for more patient-centered design. The presented research contributes substantial practical and theoretical directions for the simulation-based optimization of prosthetic vision for the blind. In parallel to clinical studies with patients and animal models, digital simulations can accelerate the cycle of hypothesis generation, prototyping and testing of implant designs. The work in this dissertation demonstrates how digital simulations with virtual reality technology, simulated prosthetic vision and deep neural networks can fulfill an important role in the interdisciplinary research and development of visual prostheses.
RkJQdWJsaXNoZXIy MTk4NDMw