Statistical signal processing for non- invasive neurological data

点击翻译按钮获取中文摘要

Biomedical Technologies (BT) is identified as a key strategic area in Canada, and developing advanced signal processing tools is vital for the advancement of BT. The amount and thus analysis complexity of non-invasive neurological data, such as scalp electroencephalography (EEG), surface electromyography (EMG) and functional magnetic resonance imaging (fMRI), is expected to increase, both with advances in medical hardware and increasing pressures from an aging population. The proposed research continues the applicant's work on statistical signal processing theory and applications, with the long term goal of developing novel, fundamental signal and information processing theory and methods that will have a lasting impact on the field of neurological data analysis. We will address key challenges in real-world problems such as assessing brain connectivity by exploring several fundamental signal processing topics: a.) efficient compression methods, with focus on sensor compression, for compressing multichannel EEG/EMG/fMRI data; b.) frameworks for automatic artifact removal/correction in fMRI, EEG and simultaneous EEG-fMRI respectively; c.) graphical models (e.g. dynamic Bayesian networks, exact Bayesian and robust LASSOs) for accurately inferring brain connectivity from neurological data incorporating key challenges, including sparsity, temporal-variability, error control and group analysis; d.) novel data-driven fusion frameworks for multimodal data analysis; e.) sensitivity analysis to study how individual elements contribute to the overall network function, and explore optimal control theory for the control/intervention of the network dynamics.