Decoding Attentional Control from Noninvasive Measures in Humans

Auditory selective attention enables us to focus on one sound within a mixture of noises. Unlike in vision, where we can easily shift attention by moving the eyes, auditory attention can only be achieved covertly via cognitive control. That is, we change our internal attentional state in the brain in order to attend to or ignore a sound object. This control is effortless and swift for healthy people, but can be challenging for people with neurological conditions such as attention de?cit/hyperactivity disorder or autism. This makes it important to study the neural mechanisms that underlie auditory selective attention. Neuroimaging technologies allow scientists to study brain activity without invasive procedures. Electroencephalography (EEG), a measure of electrical potential at the scalp, has become a popular imaging modality for neuroscience studies due to its simple setup, low cost, and high sampling rate, allowing us to capture rapid changes in electrical signals given off by the brain. EEG can thus track brain dynamics with high temporal resolution, but it is difficult to localize the location in the brain that is generating the measured activity. Another neuroimaging modality, functional magnetic resonance imaging (fMRI), measures blood oxygenation levels, which change locally when metabolic activity in a particular brain region increases. However, this change takes several seconds. fMRI signals provide millimeter level spatial resolution, but poor temporal resolution. Neither of these two modalities, nor any other noninvasive neuroimaging methods currently available, can simultaneously achieve

Effectively combining the information from EEG and fMRI could allow one to determine both when and where in the brain control of auditory attention happens. One technique for fusing neuroimaging modalities is representational similarity analysis (RSA). First, the information in brain signals is summarized via the difference among all pairs of experimental conditions, reflecting the information carried by the underlying neural representation. The resulting similarity matrix has the same dimension and scale regardless of whether it was derived from EEG or fMRI, and thus can be used to integrate information in these two neuroimaging modalities. Here I report results of an experiment that required different types of auditory selective attention (attention to space and attention to acoustic features). I collected EEG and fMRI data from healthy young adults performing these tasks. From EEG, I discovered that the event-related time course of both raw EEG voltages and alpha oscillations (8 { 14 Hz) change reliably as subjects engage in auditory attention; these show that neural representations of attentional state change as a function of time. From fMRI, I identi?ed several brain regions in the frontal (including superior and inferior precentral sulcus and inferior frontal sulcus), parietal (including intraparietal sulcus and superior parietal lobule), temporal (superior temporal gyrus) and occipital (primary visual cortex) lobe that are actively engaged during auditory attention. This allowed me to extract neural representations of attentional control as a function of location in the brain. I then conducted an RSA to fuse EEG and fMRI results and reveal the dynamics of information in different brain regions across the course of each trial. Finally, as an attempt to translate these neuroscience ?ndings into real-life applications, I explored the feasibility of decoding attention from single-trial EEG signals in order to develop an attention-based brain-computer interface (BCI) system. This dissertation identi?ed important neural signatures in EEG signals that are evoked or induced by auditory selective attention, as well as a brain network that seems to be associated with these signatures. It is among the very ?rst studies to adopt RSA fusion techniques to study information  flow through the brain during attentional control, which could be an important reference for future studies in this ?eld. Finally, this work examined several different ways to decode attention from single-trial EEG signals, achieved promising results from these attempts, and suggested possible ways to improve for future BCI development.