Dynamic features of auditory bistable perception extracted from human electrocorticography recordings

The study of dynamical systems that describe physical processes, such as a ball swinging from a rope, can be traced back to Newton's laws of motion. In the classical sense, the governing equations for a given system are typically known, and they allow for the development of strong, theoretically sound statements describing the natural world. For many modern applications in the life sciences, however, the equations describing a system may be incomplete or unknown entirely, leaving researchers reliant on observations or measurements to provide information about the underlying system. My goal as a mathematician working in neuroscience is to study neurobiological and behavioral data using data-driven methods to describe properties typically studied in dynamical systems (e.g., stable states, periodic solutions, asymptotic behaviors). In this thesis, we examine electrocorticography (ECoG) recordings from electrodes implanted on the brains of human neurosurgical patients performing a behavioral task. We integrated methodologies from dynamical systems, large-scale data analysis, and statistics to extract spatio-temporally coherent features in neural activity from high-fidelity brain recordings.

The behavioral task we study examines perceptual multi-stability, a phenomenon in which an observer's perception of a stimulus can switch between two or more interpretations called perceptual states. We invoke an auditory streaming task that has been shown to produce spontaneous switching between perceptual states. A listener is presented with a sequence of tones with the repeated pattern ABA- where A and B represent tones fixed at a high and low frequency, respectively, and `-' represents a brief silence. For certain frequency differences between A and B tones, a listener reports spontaneous alternations between two perceptual states: 1-stream in which the stimulus is integrated into a galloping-like stream, or 2-streams in which the stimulus is perceived as two segregated streams, similar to a Morse code signal. Our goal is to study the link between the dynamic properties of the electrophysiological recordings and the concurrently recorded subject-reported perception.

The evolution of the auditory-induced signal within the brain can be strikingly complex due to the vast number of neurons and their interactions through several core mechanisms. However, recent research has indicated that many internal signals within the brain evolve on low-dimensional spaces. Based on these findings, we hypothesize that there are underlying patterns manifested in ECoG data that can characterize low-dimensional dynamics; therefore, we use methods that combine data-driven dynamical systems with dimensionality reduction techniques to decompose the underlying processes into a reduced collection of meaningful and coherent features. More traditional approaches ignore the time dependence of the ECoG data and often utilize so-called `black box' classification methods. These techniques are effective in establishing the existence of a difference in neural activity between the two perceptual streams, but they are unlikely to identify underlying neural features that give rise to mutually exclusive perceptual states. By considering the temporal dependence of the recordings, this thesis shows that perceptual alternations are modulated in part by a slow rhythm that shifts the phase of relevant frequency features present in the recorded neural activity.