Personalised Signal Processing for Cortical and Cardiac Applications

Abstract

Biomedical signals reflect alterations in human physiological parameters in both healthy and pathological conditions. Their inherent variability over time and across individuals reduces the reproducibility of results and utility of biomedical signals. Personalisation of signal processing schemes by including parameters associated with the sources of inter-session and inter-subject variability can promote the usability of biomedical signals for larger cohorts. This thesis explores strategies for personalising signal processing techniques for the assessment of cortical and cardiac electrophysiological phenomena. A sensorimotor rhythm-based brain-computer interface (BCI) exploits changes in electroencephalogram (EEG) during motor imagery tasks and can establish a direct communication link between the brain and a computer, which may augment motor performance. Dealing with the variability inherent in EEG signals is not trivial and yet to be understood comprehensively to deliver BCI technology for practical use. A waveletbased signal processing method has been applied to model inter-subject associative source activations, leading to a more generalised BCI design. Intracardiac electrograms (EGM) are important for mapping electrical activation across the heart. Multiple variables, including bipolar vector orientation relative to the wave propagation vector, inter-electrode spacing, impact EGM recording. In this thesis, intracardiac EGM recorded with a customised array of electrodes were analysed to assess the impact of bipolar vector orientation and inter-electrode spacing on atrial fibrillation mapping. A novel spatial filtering method has been proposed to reduce the measurement uncertainty due to bipolar vector orientation. Besides, an independent component analysis-based filtering has been proposed as a potential preprocessing method for eliminating ventricular far-field artefact.