8.5. Limitations and future directions 8 121 first step in addressing these socio-ethical concerns is the awareness among researchers, clinicians and engineers. 8.5. Limitations and future directions Besides the specific limitations described throughout the chapters of this dissertation, several more general limitations can be identified. First and foremost: despite efforts to narrow the gap between simulations and reality, this dissertation makes use of idealized models of reality: convolutional neural networks are not identical to the visual system; virtual reality depicts only a virtual reality; sighted study participants behave differently from blind individuals and -most importantly- our phosphene simulations are still characterized by some crucial reality gaps. Nevertheless, it is vital to acknowledge that models will always be (inherently) different from reality. Secondly, the experimental work in this dissertation is the product of many arbitrary choices, such as the choice of environment and the experimental instructions. This hinders the ability to draw comparisons with other studies. This lack of standardized testing forms a general challenge in the field. It is difficult to measure the effectiveness of visual prostheses using clinical vision tests (Peli, 2020) and the outcomes have a limited practical relevance. Although there is still no agreement on the best way to report outcomes of visual prostheses (Erickson-Davis & Korzybska, 2021), efforts have been undertaken to design standardized, ‘real world’ experimental tasks (e.g., Bach et al., 2010; Geruschat et al., 2015). These are valuable directions for further exploration, as future simulation research would benefit from such standardization. Thirdly, the benefits of the computational models presented in this dissertation have not been validated in an experimental setting. A pitfall of computational models is that without the right constraints they can be naive to human requirements. Besides the computational optimization of information encoding ‘effectiveness’, it is important to evaluate the intuitive informativity for human observers. Note that interesting directions are being explored in other work, such as more brain-like deep learning models (Schrimpf et al., 2020) or optimization frameworks that include patients in the optimization loop (Granley et al., 2023). Nevertheless, computational frameworks should be regarded as a complementary addition, and not a replacement of behavioral and clinical work. 8.6. Conclusion The past decades have revealed promising progress in the development of visual neuroprostheses that supply a functional form of visual perception to individuals who became blind. Nevertheless there are many remaining challenges regarding the encoding of visual signals into the brain and the functional outcomes depend on many hardware, software or contextual parameters.Inspired by the scientific and technological developments in the fields of biomedical engineering, neuroscience and artificial intelligence, this dissertation explores the value of digital simulations for the optimization of visual prostheses. In particular, the research in this dissertation uses virtual reality technology, simulated prosthetic vision and deep neural networks for prototyping and optimization of cortical visual prosthetics.
RkJQdWJsaXNoZXIy MTk4NDMw