More than 40 million people worldwide are blind. Scientists are working to develop implants in the brain that will allow blind people to see a little, for example a pattern of a streetcar or a cyclist. The PhD research of Jaap de Ruyter de Steveninck of Radboud University is helping with this development by using AI and deep learning.
’Visual prostheses are a promising development,’ the neuroscientist explains. ’With electrical stimulation in the visual brain regions, a very basic form of vision can be restored in people with blindness.’ In practice, this amounts to a camera on the blind person’s body that records the environment. Intelligent software simplifies the environment with AI, after which the brain of the blind person is stimulated in such a way that dots of light can be "seen". At the moment these are only a few dots, but in the future there may be enough of them to show, for example, the outline of objects: hopefully a streetcar, or a cyclist.
Deep learning
In his research, De Ruyter de Steveninck used VR simulations and deep learning to look at the usability of this basic vision: does it really help people who can’t see? He also looked at how the simulations could be optimized. ’We looked at what happened with many bright dots - for which you also need many electrodes in the brain - and with fewer dots. We also gave sighted people VR glasses with such light point simulations and instructed them to walk through a room with obstacles.’ The researcher looked at which algorithms worked best to get the environmental representation as good as possible. He also looked at whether testing with people could be automated by using AI. It turned out that it could. ’AI can be used to test these simulations and thus offers opportunities for faster optimization. People as test subjects remain important, but this could make the process of creating the best simulation a lot easier and faster in the future.’
Small amount of vision
With the results of his research, De Ruyter de Steveninck has developed digital simulations and software that will take the development of intelligent neuroprosthetics a step further. ’A very small amount of vision can already help people who can’t see. It will take time before the technology is so far advanced that blind people can really see something and there are still many factors unknown. But this research makes an important contribution.’
