Facilitating the incorporation of neuroscience methods and knowledge into brain-computer interfaces

Thesis (Ph.D.)--University of Washington, 2017-03This body of work is focused on brain-computer interfaces (BCIs) – devices that interpret brain activity in order to generate artificial output. BCIs have a great deal of promise in both rehabilitative and commercial domains. However, there is still a list of obstacles that must be solved before BCIs can advance out of controlled research settings and into real world scenarios. Some obstacles arise due to a growing separation between neuroengineering and neuroscience; many modern neuroscience publications report findings at the cortical level. Therefore, it is difficult to leverage this basic science research in BCIs that use non-invasive recordings (e.g., electroencephalography or magnetoencephalography) because those data are recorded from outside the head. To alleviate this disconnect, we borrow a technique from neuroimaging called “source imaging,” which allows the estimation cortical activity (on the surface of the brain) from non-invasive data (recorded on or above the surface of the scalp). We used source imaging in an attempt to address three separate issues facing the BCI field. First, we showed that targeting activity from a specific region of cortex known to be important for auditory spatial attention led to improved performance, compared to a traditional sensor-based approach, when classifying if a subject switched attention. Second, we developed a new transfer learning technique aimed at reducing the 20-30 minute calibration period required for most non-invasive BCIs. This method used source imaging to partially normalize variance due to head anatomy, brain anatomy, and electrode positioning, and could train a subject-independent BCI with higher accuracy than the traditional subject-dependent model despite using no training data from the subject of interest. Third, we explored functional connectivity to generate features for a BCI attempting to classify resting state. While performance did not surpass traditional methods, we outline future directions and the importance of continued research. Overall, these three aims are focused on using tools and research findings from neuroscience as a principled way to advance BCI methodology