ABSTRACT OF THE DISSERTATION

The role of alpha oscillations in visual information processing

by

Stephanie Nelli

Doctor of Philosophy in Neurosciences with a Specialization in Computational Neuroscience

University of California San Diego, 2019

Professor John Serences, Chair

Neural oscillations are one the most prominent features of electrical brain recordings, involving the synchronized activity of large populations of neurons, and have been linked to a variety of important functions over the past century. Existing theories propose these oscillations allow the brain to dynamically switch between functional neural circuits, allowing computational flexibility on time scales too fast for the slow structural changes that characterize long-term cortical plasticity. In particular, alpha oscillations (~8-13 Hz) are often visible in raw scalp electroencephalography (EEG) recordings over visual and parietal cortex and seem to regulate the waxing and waning of visual attention and perception across time. However, many previous results focus specifically on the phase, power or frequency of the alpha oscillation, neglecting the common dynamical systems that simultaneously generate all these metrics and impact visual information processing. In this dissertation, I begin by reviewing relevant literature about alpha oscillations, and then aim to link together disparate measures of the alpha oscillations in both behavior and the brain to address how dynamical alpha state regulates visual information processing. First, in C1 I find that both phase locked and purely power based analyses of imprecise attentional selection display alpha rhythms, suggesting a common impact on behavior. Indeed, in C2 I show that alpha frequency, which governs phase and amplitude are linked mathematically in simple models of harmonic oscillators, which I confirm in neural recordings. Thus, the impact of alpha oscillations on perception depend on circuit interactions with top-down driving oscillators, and in C3 I find that optimal oscillatory drive for visual perception depends intimately on each subject’s particular dynamical system and resultant peak alpha frequency. Together, this thesis challenges core assumptions underlying current theories of the role that alpha oscillations play in regulating visual information processing using both mathematical models and empirical data. I then propose a more unified theoretical framework in which alpha frequency, phase and amplitude should not be viewed as independent metrics to be correlated with behavior, but instead as the result of a common dynamical system that impacts visual perception.

In a complex visual environment—such as a crowed street in Bangkok— driving would be impossible if drivers do not have an intact attentional system. They have to monitor cars and trucks surrounding their vehicles, and at the same time they have to also attend to traffic lights and street signs and watch out for pedestrians and motorbikes that could cross the street unexpectedly at any given time. In this type of scenarios, drivers need to divide their attention into multiple spotlights and flexibly change the size of their attention field (zoom-in and zoom-out) such that relevant information is most efficiently encoded at the expense of irrelevant information. The first two experiments in my dissertation examined neural mechanisms underlying these two natures of our attentonal system, and the last experiment officially evaluated the relative contributions of alternative neural mechanisms that may account for attention-related improvement in perceptual performance. In the first experiment (Chapter 2), we provided neural evidence showing that attention could be divided into multiple spotlights across non-contiguous locations in the visual scene, even when visual objects were in close proximity (i.e., in a single quadrant). Using a stimulus- frequency-tagging technique where we flashed two visual targets and a distractor at the intermediate location at different frequencies, we were able to monitor changes of steady state visually evoked potentials (SSVEPs) that oscillated at the same frequencies as the target and distractor stimuli. We found the significant divergence of the target-related and distractor-related SSVEPs ~150-350ms before human participants correctly discriminated the orientations of the two targets. In the second experiment (Chapter 3), we examined the neural basis underlying changes of the spatial scope of attention and studied how such changes may alter the way sensory information is encoded in the visual cortex. By manipulating the spatial extent of visual target in a stream of flickering non-target stimuli, we observed changes in the spread of cortical activity in the contralateral visual cortex measured using functional magnetic resonance imaging (fMRI). As attention became more distributed due to the uncertainty of target locations, we observed a larger spread of cortical activity compared to when attention was more focused to a single target location. Importantly, we found that this spread of the spatial attention modulated the magnitude of attentional modulations of sensory signals measured via SSVEPs in the way that was consistent with predictions from computational models based on divisive normalization. Lastly, in the third experiment (Chapter 4), we made a further step to formally examine the quantitative relationship between attentional gain modulations of neural signals and attention-related improvement in behavioral performance, and evaluated the relative contributions of attentional gain mechanisms and other alternative mechanisms, including noise reduction and efficient read-out mechanisms. We found that in a relatively simple attention task, attentional gain modulations of early visually evoked responses measured via electroencephalography (EEG) could sufficiently predict attention-related improvement in perceptual performance, without the need to invoke the other alternative mechanisms. Taken together, the results from these three experiments suggest that selective attention enhances sensory information processing via changes in gain modulations of early sensory signals and these attentional gain modulations play a critical role in supporting attention-related improvement in perceptual performance.

Our visual world is full of ambiguous sensory signals, from which we have to extract relevant and meaningful information in order to guide optimal actions. To maximize the efficiency of this process, our visual system relies on foreknowledge to prioritize the processing of relevant or expected features. Knowledge of statistical regularities in the environment can lead to faster detection and recognition of objects when they are encountered in an expected context (e.g., a bird in a backyard) than when they are encountered in unlikely context (e.g., a bird in a washing machine). In addition, knowledge about the current task goals can also support faster and more accurate processing of relevant over irrelevant items--a mechanism referred to as selective attention. In what manner do these “top down” modulatory factors individually and jointly affect visual sensory processing, decision making, and behavior? In three studies, we examined how perceptual decision making is modulated by prior expectation about stimulus probabilities alone and in the context where knowledge about the current behavioral goals were available. We examined these effects both neurally via electroencephalography (EEG) and behaviorally through psychophysics and also in amnesic patients in relation to age-matched controls. To this end, we first devised an experimental paradigm where prior expectation and selective attention could be individually manipulated. The behavioral readouts from this paradigm were continuous which made it possible for the temporal evolution of the effects of expectation and attention on decision process to be probed both behaviorally and in relation to the continuous neural (EEG) measures. We first demonstrated that prior expectation improves decision processes by primarily affecting post-perceptual operations such as initiation and execution of motor responses, instead of directly improving the efficiency of early sensory processing. This finding confirms an idea that has been put forth by traditional theoretical framework that prior expectation affects decision making by preferentially modulating motor responses that correspond to sensory inputs with high probability of occurring. Further, we showed that while both expectation and attention improved behavior, the underlying neural mechanisms that give rise to these effects differed: while attention operates on the early processing of sensory inputs, expectation affects the late stage of decision making by biasing motor responses towards the most likely decision choice. These differential temporal dynamics of expectation and attention were observed bot h behaviorally and neurally. Finally, we demonstrated that an ability to utilize knowledge about current task goals and to form expectation based on statistical regularities of the sensory environment can be independent of a declarative memory system.

Selective attention in vision undoubtedly uses many different types of mechanisms to achieve better processing of behaviorally relevant visual information. The present dissertation inspects two such possible mechanisms. The first proposed mechanism is through selective spatial attention. We proposed that “attentional control” regions of the brain send selective attention signals that relate to the differential attentional modulation measured in visual cortex, corresponding to the attended and ignored sides of visual space. We find, surprisingly, that the BOLD signal of attentional control regions do not appear to code for selective spatial attention in this manner. fMRI signals therefore appear to prioritize general arousal effects of attention more so than the types of attentional control signals that frontoparietal regions send to early visual cortex. The second proposed mechanism is through sampling visual information through time. It is hypothesized that intrinsic neural oscillations represent inhibitory waves of cortical processing ability, in which subjects are more or less able to detect visual stimuli depending on the phase of α oscillations in the brain. We used a paradigm involving visual entrainment; this paradigm has been used in the past to support the “pulsed inhibition of α” hypothesis. We find that the behavioral effects of this paradigm cannot strongly be supported by the pulsed inhibition of α because the neural effects of visual entrainment are transient and do not carry over into the target period.

Attention—the process of selecting and prioritizing relevant stimuli in our environment—has long been considered a highly important cognitive process that enables us to deal with the large volume of information that we encounter at any moment. While we still do not have a complete understanding of the mechanisms that guide selection, different theories have carved out how attention might function with respect to distinct stimulus domains, namely spatial locations, bound objects, or basic visual features. A large body of research has emerged that investigates the behavioral implications of attentional selection within these domains, as well as the neural processes that underlie this selection. To extend upon this work, in this dissertation I present a series of studies exploring fundamental aspects of feature-based attention: how attention spreads between different features within a single object, as well as globally across locations in the visual field; how broader and narrower ranges of relevant colors can be attended simultaneously and efficiently; how attention warps the perception of colors in service of efficient processing, and; how information about target and distractor colors are carried within a neural marker of visual processing, and how representations of those colors are affected by attention. In each of these studies, I tie the main findings back to broader theories connecting feature-based attention with object-based and spatial theories of attention. By considering how these theories interact and make claims about the nature of attentional selection, I argue for a framework in which attention is thought of as a process by which the representational content of information is shaped in support of adaptive behavior, as opposed to a mechanism which highlights particular types of stimuli.