Algorithms for enhanced spatiotemporal imaging of human brain function

Abstract

Studies of human brain function require technologies to non-invasively image neuronal dynamics with high spatiotemporal resolution. The electroencephalogram (EEG) and magnetoencephalogram (MEG) measure neuronal activity with high temporal resolution, and provide clinically accessible signatures of brain states. However, they have limited spatial resolution for regional dynamics. Combinations of M/EEG with functional and anatomical magnetic resonance imaging (MRI) can enable jointly high temporal and spatial resolution. In this thesis, we address two critical challenges limiting multimodal imaging studies of spatiotemporal brain dynamics. First, simultaneous EEG-fMRI offers a promising means to relate rapidly evolving EEG signatures with slower regional dynamics measured on fMRI. However, the potential of this technique is undermined by MRI-related ballistocardiogram artifacts that corrupt the EEG. We identify a harmonic basis for these artifacts, develop a local likelihood estimation algorithm to remove them, and demonstrate enhanced recovery of oscillatory and evoked EEG dynamics in the MRI scanner. Second, M/EEG source imaging offers a means to characterize rapidly evolving regional dynamics within an estimation framework informed by anatomical MRI. However, existing approaches are limited to cortical structures. Crucial dynamics in subcortical structures, which generate weaker M/EEG signals, are largely unexplored. We identify robust distinctions in M/EEG field patterns arising from subcortical and cortical structures, and develop a hierarchical subspace pursuit algorithm to estimate neural currents in subcortical structures. We validate efficacy for recovering thalamic and brainstem contributions in simulated and experimental studies. These results establish the feasibility of using non-invasive M/EEG measurements to estimate millisecond-scale dynamics involving subcortical structures. Finally, we illustrate the potential of these techniques for novel studies in cognitive and clinical neuroscience. Within an EEG-fMRI study of auditory stimulus processing under propofol anesthesia, we observed EEG signatures accompanying distinct changes in thalamocortical dynamics at loss of consciousness and subsequently, at deeper levels of anesthesia. These results suggest neurophysiologic correlates to better interpret clinical EEG signatures demarcating brain dynamics under anesthesia. Overall, the algorithms developed in this thesis provide novel opportunities to non-invasively relate fast timescale measures of neuronal activity with their underlying regional brain dynamics, thus paving a way for enhanced spatiotemporal imaging of human brain function.