Date of Award
Doctor of Philosophy (PhD)
Introduction: Brain Computer Interfaces (BCIs) have been shown to restore lost motor function that occurs in stroke using electrophysiological signals. However, little evidence exists for the use of BCIs to restore non-motor stroke deficits, such as the attention deficits seen in hemineglect. Attention is a cognitive function that selects objects or ideas for further neural processing, presumably to facilitate optimal behavior. Developing BCIs for attention is different from developing motor BCIs because attention networks in the brain are more distributed and associative than motor networks. For example, hemineglect patients have reduced levels of arousal, which exacerbates their attentional deficits. More generally, attention is a state of high arousal and salient conscious experience. Current models of consciousness suggest that both slow wave sleep and Propofol-induced unconsciousness lie at one end of the consciousness spectrum, while attentive states lie at the other end. Accordingly, investigating the electrophysiology underlying attention and the extremes of consciousness will further the development of attentional BCIs.
Phase amplitude coupling (PAC) of neural oscillations has been suggested as a mechanism for organizing local and global brain activity across regions. While evidence suggests that delta-high-gamma PAC, which includes very low frequencies (i.e. delta, 1-3 Hz) coupled with very high frequencies (i.e. gamma 70-150 Hz), is implicated in attention, less evidence exists for the involvement of coupled mid-range frequencies (i.e. theta, 4-7Hz, alpha: 8-15 Hz, beta: 15-30 Hz and low-gamma: 30-50 Hz, aka TABL PAC). We found that TABL PAC correlates with reaction time in an attention task. These mid-range frequencies are important because they can be used in non-invasive electroencephalography (EEG) BCI’s. Therefore, we investigated the origins of these mid-frequency interactions in both attention and consciousness. In this work, we evaluate the relationship between PAC to attention and arousal, with emphasis on developing control signals for an attentional BCI.
Objective: To understand how PAC facilitates attention and arousal for building BCI’s that restore lost attentional function. More generally, our objective was to discover and understand potential control features for BCIs that enhance attention and conscious experience.
Methods: We used four electrophysiological datasets in human subjects. The first dataset included six subjects with invasive ECoG recordings while subjects engaged in a Posner cued spatial attention task. The second dataset included five subjects with ECoG recordings during sleep and awake states. The third dataset included 6 subjects with invasively monitored ECoG during induction and emergence from Propofol anesthesia. We validated findings from the second dataset with an EEG dataset that included 39 subjects with EEG and sleep scoring.
We developed custom, wavelet-based, signal processing algorithms designed to optimally calculate differences in mid-frequency-range (i.e. TABL) PAC and compare them to DH PAC across different attentional and conscious states. We developed non-parametric cluster-based permutation tests to infer statistical significance while minimizing the false-positive rate. In the attention experiment, we used the location of cued spatial stimuli and reaction time (RT) as markers of attention. We defined stimulus-related and behaviorally-related cortical sites and compared their relative PAC magnitudes. In the sleep dataset, we compared PAC across sleep states (e.g. Wake vs Slow Wave Sleep). In the anesthesia dataset, we compared the beginning and ending of induction and emergence (e.g. Wake vs Propofol Induced Loss of Consciousness)
Results: We found different patterns of activity represented by TABL PAC and DH PAC in both attention and sleep datasets. First, during a spatial attention task TABL PAC robustly predicted whether a subject would respond quickly or slowly. TABL PAC maintained a consistent phase-preference across all cortical sites and was strongest in behaviorally-relevant cortical sites. In contrast, DH PAC represented the location of attention in spatially-relevant cortical sites. Furthermore, we discovered that sharp waves caused TABL PAC. These sharp waves appeared to be transient beta (50ms) waves that occurred at ~140 ms intervals, corresponding to a theta oscillation. In the arousal dataset DH PAC increased in both slow wave sleep (SWS) and Propofol-induced loss of consciousness (PILOC) states. However, TABL PAC increased only during PILOC and decreased during SWS, when compared to waking states. We provide evidence that TABL PAC represents “gating by inhibition” in the human brain.
Conclusions: Our goal was to develop electrophysiological signals representing attention and to understand how these features explain the relationship between attention and low-arousal states. We found a novel biomarker, TABL PAC, that predicted non-spatial aspects of attention and discriminated between two states of unconsciousness. The evidence suggested that TABL PAC represents inhibitory activity that filters out irrelevant information in attention tasks. This inhibitory mechanism of was confirmed by significant increases in TABL PAC during Propofol anesthesia, when compared to SWS or waking brain activity. We conclude that TABL PAC informs the development of electrophysiological control signals for attention and the discrimination of unconscious states.
Eric C. Leuthardt
Dan Moran, Barani Raman, Linda Larson-Prior, ShiNung Ching,