Current State and Future of Brain-Machine Interfaces
In this presentation Miriam Shinichi, an assistant professor at USC Electrical Engineering and neuroscience
talks about brain machine interfaces.
Our brain consists of 86 billion neurons and it's the
activity of these neurons that represent our emotions thoughts and action in particular not only are our brains function
such as movements represented in neural activity what's or is dysfunctions, for example in neuropsychiatric disorders
like depression or even in induced states such as anesthesia or coma. So the question is what if we can actually build a
direct interface of the brain and record the neural activity how can we build brain machine interfaces that can help us
treat neurological disorder or even understand how the brain works better. And of course there's many challenges that we
have to overcome before we can have a brain machine interface.
The first challenge obviously is to have a technology that
allows us to record brain activity and you can think of these recordings at various levels of invasiveness and spatial temporal time scale, for example a very non-invasive modality that you can think
about is an electrode cap that you can wear that records a modality that we call EEG but this is going to have very low signal to noise ratio are not enough spatial resolution so then on the other extreme of invasiveness you can imagine micro electrode arrays that penetrate the cortex to give you the finest spatial resolution so that you can actually record the activity at the level of single neurons an activity that we call the sparking activity that represents the presence or absence of an action potential and because it's a binary event they usually represented as a binary time series, and there's modalities of course in-between but it does don't just need a recording technology we also need to be able to writing information potentially by having a stimulation technology either electrically or optically so let's say
we have these technologies and we're going to get a ton of data from very high dimensional brain networks and we have to make sense of it so the next challenge would be how do you use these these that data that will collect from the brain to infer a state of the brain and because we want to build a technology that interacts with the brain how can I use methods from control theory to control brain state so you can have tools from machine learning statistical inference and control theory brought into the field to develop these types of algorithms. And then finally once we have these algorithms of
course you have to have an implantable device and this device has to be possibly violent low power efficient and by compatible so there's a lot of challenges on the side of integrated circuits that people are working on so as you can see this is a truly interdisciplinary field at the interface of neuroscience neurosurgery, electrical engineering,
statistical inference control biomedical engineering and really that the field is quite diverse in the sense of people with various expertise that try to build these systems.