Tensor-based blind source separation for structured EEG-fMRI data fusion

A complex physical system like the human brain can only be comprehended by the use of a combination of various medical imaging techniques, each of which shed light on only a specific aspect of the neural processes that take place beneath the skull. Electroencephalography (EEG) and functional magnetic resonance (fMRI) are two such modalities, which enable the study of brain (dys)function. While the EEG is measured with a limited set of scalp electrodes which record rapid electrical changes resulting from neural activity, fMRI offers a superior spatial resolution at the expense of only picking up slow fluctuations of oxygen concentration that takes place near active brain cells. Hence, combining these very complementary modalities is an appealing, but complicated task due to their heterogeneous nature. In this thesis, we devise advanced signal processing techniques which integrate the multimodal data stemming from simultaneously recorded EEG and fMRI. We focus on their application in refractory epilepsy, wherein some brain cells undergo hypersynchronous activity, leading to seizures that cannot be suppressed by medication. In such cases, EEG?fMRI can aid a presurgical evaluation of the patient, to infer the location in the brain where epileptic discharges originate. In the first part of the thesis, we propose a semi-automatic technique which drastically improves the EEG signal quality, based only on a few examples of recorded epileptic discharges. We use this output to inform an fMRI analysis which searches the brain volume for correlated fluctuations, and demonstrate that this approach is more sensitive than existing methods in detecting fMRI signal changes in epileptic brain regions. In the second part, we develop data fusion approaches based on representations of the data as tensors (?multidimensional arrays?) to capture the rich, complex nature of EEG and fMRI, and to exploit their attractive properties for data mining. We perform simultaneous blind signal separation of both data tensors into distinct ?components?, each representing a source of variation in the data. Our experiments show that these customized coupled tensor decompositions are not only able to extract components that model the temporal, spatial and spectral profiles of epileptic discharges, but also to estimate the variable functional relationship between EEG and fMRI, i.e., neurovascular coupling. Clinical validation shows that these novel techniques produce complementary sets of biomarkers, which assist the characterization and diagnosis of epilepsy.