Retinal dystrophies such as retinitis pigmentosa and macular degeneration induce progressive loss of photoreceptors, resulting in profound visual impairment in more than ten million people worldwide. Visual neuroprostheses (?bionic eyes?) aim to restore functional vision by electrically stimulating remaining cells in the retina, analogous to cochlear implants. A wide variety of neuroprostheses are either in development (e.g. optogenetics, cortical) or are being implanted in patients (e.g. subretinal or epiretinal electrical). A limiting factor that affects all device types are perceptual distortions and subsequent loss of information, caused by interactions between the implant technology and the underlying neurophysiology. Understanding the causes of these distortions and finding ways to alleviate them is critically important to the success of current and future sight restoration technologies. In this proposal, human visual psychophysics, computational modeling, data-driven approaches, and virtual reality (VR) will be combined to develop and experimentally validate optimized stimulation protocols for epiretinal prostheses. This approach is analogous to virtual prototyping for airplanes and other complex systems: to use a high-quality model of both the implant electronics and the visual system in order to generate a ?virtual patient?. Retinal electrophysiological and visual behavioral data will be used to develop and validate a computational model of the expected visual experience of patients when electrically stimulated. One way of using this model will be to generate simulations of the expected perceptual outcome of electrical stimulation across a wide variety of electrical stimulation patterns. These will be used as a training set for machine learning algorithms that will invert the input-output function of the model to find the electrical stimulation protocol that best replicates any desired perceptual experience. The model can also be used to simulate the expected perceptual experience of real patients by using sighted subjects in a VR environment ? ?VR virtual patients?. These virtual patients will be used to discover preprocessing methods (e.g., edge enhancement, retargeting, decluttering) that improve behavioral performance in VR. Although current retinal prostheses have been implanted in over 250 patients worldwide, experimentation with improved stimulation protocols remains challenging and expensive. Implementing ?virtual patients? in VR offers an affordable and practical alternative for high-throughput experiments to test new stimulation protocols. Stimulation protocols that result in good VR performance will be experimentally validated in real prosthesis patients in collaboration with Second Sight Medical Products Inc. and Pixium Vision, two leading device manufacturers in the field. This work has the potential to significantly improve the effectiveness of visual neuroprostheses as a treatment option for individuals suffering from blinding retinal diseases.
Inadequate stimulation paradigms are currently one of the main factors limiting the effectiveness of visual prostheses as a treatment option for individuals suffering from blinding retinal diseases. My goal is to develop and validate novel stimulation protocols for visual prosthesis patients that minimize perceptual distortions and thereby improve behavioral performance. Developing methods for generating better stimulation protocols through a combination of behavioral testing, virtual reality, computational modeling, and machine learning, has the potential to provide a transformative improvement of this device technology.
