Development of a More Robust Near-Infrared Spectroscopy Brain Computer Interface: Single-Trial Classification of Near-Infrared Spectroscopy Signals Arising from Multiple Cortical Regions

Near-infrared spectroscopy (NIRS) has been gaining increasing attention as a brain-computer interface (BCI) modality. However, communication rates offered by this technology are not yet practical for clinical implementation. This thesis investigates the feasibility of enhancing the speed and accuracy attainable by NIRS-BCIs. To date, most studies have utilized activity from a single cortical region, particularly the anterior prefrontal cortex (aPFC), to discern two non-motor-based mental states. However, cognitive processes require coordination between several, spatially-distributed areas. Additional cortical information may help improve classification of mental states, particularly combinations of more than two. Furthermore, testing has largely focused on offline analyses, despite the possibility of enhanced cortical control attainable with feedback. Lastly, NIRS-BCI work has primarily involved able-bodied participants, most of which were offline. In an initial offline study, we investigated the value of utilizing cortical activity measured from multiple brain regions to differentiate all possible combinations of two cognitive tasks and rest. With prefrontal and parietal measurements, average binary and ternary accuracies exceeded 80% and 70%, respectively. Accuracies and information transfer rates (ITRs) improved significantly with the inclusion of measurements beyond the aPFC. A second study was conducted to evaluate ternary classification accuracies achievable in an online system that offered classification feedback. An average online accuracy of 74.2 Âą 14.8% was attained, representing further improvements in NIRS-BCI speed and accuracy. In a final study, we tested an online system developed for a candidate BCI user with severe, congenital muscular impairments. The binary system made use of prefrontal measurements beyond the aPFC and was tailored for the user prior to each online session. Offline and online accuracies of 66.9 Âą 9.11% and 63.33%, respectively, were achieved. Collectively, these findings demonstrate the value of considering multiple brain regions and offering online classification feedback to realize a faster and more accurate NIRS-BCI.Ph.D.2018-11-07 00:00:0