Doctoral Thesis Defense: The Physical Paradigm for Bidirectional Brain-Computer Interfaces

Abstract

This dissertation deepens research into interfaces that supplement input the user transmits to the computer intentionally with an auxiliary channel describing ongoing brain activation. Existing implementations of such implicit brain-computer interfaces (BCI) depend on machine learning algorithms trained to distinguish physiological signals detected with functional near-infrared spectroscopy (fNIRS) under different task conditions, which subsequent chapters will refer to as neuracles. When calibrated to the user's brain, the implicit BCI adjusts settings in the interface to better match the mental state that has unfolded. Because this approach does not depend on an understanding about the relationship between fNIRS signals and physical activity in the brain, I will refer to the current methodology for studying and building implicit BCIs as the agnostic paradigm. Experiments evaluating implicit BCIs in the agnostic paradigm have led to measurable improvements in user performance in a number of controlled laboratory experiments.

This dissertation introduces a descendant of implicit BCI, referred to as a bidirectional BCI. Instead of adapting the interface to match the mental state that has unfolded, a bidirectional BCI strives to adapt outputs to the brain to stimulate and maintain optimal mental states for its user. This new class of BCI depends on discovering a model for the interaction between brain and computer at four levels of analysis. Such a model should account for how the brain works at the physical level, the linkup between brain state and mental state at a mental level, the relationship between brain state and sensor data at the neuracle level, as well as how computer settings and output affect the physical state of the brain at an interface level. With a synchronized model at these four levels, a bidirectional BCI can establish a feedback loop between the user's brain and its methods to affect the brain's state, and deploy machine learning algorithms to adjust output to the brain to coerce and sustain desirable mental states.

But bidirectional BCIs are not possible with the existing agnostic paradigm. This dissertation therefore develops an alternative method, which has a synchronized understanding of brain-computer interaction at physical, mental, neuracle, and interface levels. Scientific progress towards physical neuracles depends on methods for studying one's own brain as a scientist and engineer, as well as other lucid individuals synchronized on a common vocabulary for describing mental states. This alternative methodology is facilitated by the Neuracle software distributed as part of this dissertation, which consists of interfaces, visualizations, and signal processing algorithms needed in a physical paradigm for BCI.