Modeling temporal dependency of brain responses to rapidly presented stimuli in ERP based BCIs

Abstract

Brain computer interface technologies aim, among a number of possible goals, to help people with motor and speech disabilities to communicate through a computer using the electrical activity in the brain, allowing them to move robotic arms or typing. We will specifically work with the RSVP-KeyboardT M , a letter-by-letter brain computer interface (BCI) typing system based on EEG responses to serial presentation of characters on the screen. Target stimulus elicits a P300 ERP, a positive deflection in the scalp voltage around 300 ms after the stimulus. Previous RSVP-KeyboardT M models approached the problem of detecting the user’s intended symbol by fusing a language model with EEG evidence in response to each symbol presentation, with the assumption that EEG for each trial was independent from others. Trials were windowed assuming a limited time effect of the ERPs (500ms). Even though the inter stimulus interval (ISI) is smaller than the window duration (∼200ms), independence of overlapping ERP responses from consecutive trials given the intended symbol was assumed. Our main objective here is to design a more realistic model that better captures the temporal dependencies inside a sequence. The new model consists of a finite state machine whose states are determined by the presence or nonpresence of a target within the assumed stimulus response time. Thus, the new method looks for the most probable sequence of states given the position of the target, considering the temporal dependencies induced by overlapping ERP responses on the recorded EEG. Acknowledgements I would like to thank Dana Brooks for always being supportive, comprehensive and helpful in any situation and for the big opportunity that coming to Northeastern University has been for me. Thanks to Ferran Marques as our advisor in Barcelona and for all his help and comprehension when dealing with administrative problems. I also want to dedicate an special thank to Jaume Coll for his giant patience, and for being always ready to help. And Biel Roig, who after working with me for the last two years in UPC, crossed the Atlantic ocean with me and has been sharing this awesome experience during last 10 months. This project would not have been possible without the people in CSL group. I would like to thank Deniz Erdogmus for letting me work in the BCI system and Umut Orhan for all his time working with me, his patience, help and advice. I also want to thank Murat Akcakaya and Mohammad Moghadamfalahi for their collaboration with ideas while developing the system. Also thanks for their time and enthusiasm to all the subjects that offered to participate in the experiments. Thanks also to all the friends here and back in Barcelona with who I know I can always count with despite of the distance. Finally, and most important, I would like to thank the support from my parents in all my decisions, not only during this year, but also during my whole life.