Adaptive Time-Varying Brain Source Localization

A Master of Science thesis in Biomedical Engineering by Sajedah A. Al-Momani entitled, “Adaptive Time-Varying Brain Source Localization”, submitted in November 2019. Thesis advisor is Dr. Hasan Mir and thesis co-advisor is Dr. Hasan Al-Nashash. Soft copy is available (Thesis, Approval Signatures, Completion Certificate, and AUS Archives Consent Form).

Abstract

According to the most recent statistics provided by the World Health Organization, 50 million individuals around the world are diagnosed with epilepsy with 2.5 million new cases diagnosed annually. One-third of this large population are refractory epileptic patients who resist the antiepileptic drugs. Accordingly, the recommended solution is to surgically resect the seizure onset zone (SOZ). To ensure favourable surgical outcomes in terms of being seizure free and avoiding undesirable consequences, a precise identification of the SOZ is a critical factor. Electroencephalograph (EEG) is usually used along with advanced imaging techniques to localize and identify the SOZ. The high temporal resolution of EEG makes it a convenient tool to study the dynamic and rapid propagating nature of epileptic spikes. Incorporating the spatiotemporal propagating characteristic of epileptic spikes in the formulation of the EEG source localization problem enhances the identification of SOZs corresponding to both fixed and moving sources. Therefore, the main objective of this thesis is to localize and track, with acceptable computational time and localization error, the spatiotemporal dynamics of epileptic sources. To do so, the time-varying source localization problem is solved using two methods, namely the Source Affine Image Reconstruction (SAFFIRE) algorithm and a proposed Steepest Descent algorithm. The derivation of these methods is based on finding filter weights that minimize the mean squared error cost function. The main reason behind proposing the Steepest Descent method is to obtain a new estimation of the source vector with each newly received data sample. The performance of these methods is investigated on simulated data mimicking two scenarios: a spatially fixed source and a spatially moving source. For the moving source scenario, the results showed that the SAFFIRE algorithm had a 0.9±0.05 𝑐𝑚 localization error compared to a 1.69±0.06 𝑐𝑚 for the Steepest Descent. The execution time of the SAFFIRE algorithm was 9-folds higher compared to Steepest Descent. Finally, the two algorithms are applied on a high-density simulated epileptic spike originated from a fixed source. Results showed that both algorithms provided similar localization error. This suggests that both algorithms may perform similar to each other at higher number of electrodes.