Automated Focal Seizure Localization by Joint Processing EEG and MRI Data of Epileptic Patients

Localizing epilepsy in the brain has attracted a lot of attention in diagnosing and treating epileptic seizures and is currently a major challenge. Existing methods of epilepsy localizers heavily rely on the physician expertise to estimate the location by observing the EEG, and recently the fMRI. No automated method for this diagnosis exists. This paper proposes an innovative method of combining the mutual information of EEG and MRI to develop an automated epilepsy source locator. The proposed method, denoted by ULeV-EsLORETA, includes three components: Multimodal Denoiser (ULeVMSEE), Forward Model, and Efficient High Resolution sLORETA (EHR-sLORETA). The ULeVMSEE algorithm is based on two methods: Unified Left Eigenvectors (ULEV) and Mean Squared Eigenvalues Error (MSEE). ULEV has the ability to extract the mutual information of EEG and MRI and separates the desired mutual information from the additive disturbance of the recorded data. Additionally, MSEE is employed to estimate the number of common bases based on EEG. As pre-processing methods for ULEV, standardization and resizing techniques are taken into consideration, to ensure that the multimodal datasets are compatible with one another and may be processed simultaneously. Following ULeVMSEE, the extracted bases are used to estimate the noise-free EEG using a forward model. At the end, Efficient High Resolution sLORETA (EHR-sLORETA) is utilized to extract the location of seizure sources. ULeV-EsLORETA is compared with the existing source locator methods for both synthetic and real data. The method illustrated solid advantages over the existing methods in the sense of Average Correlation Coefficient (ACC) and Average Error Estimation (AEE) and estimation of the true number. of sources. It is shown that ACC is improved on average by 17.6% and 40% for different number of common sources and various SNR-difference respectively. Additionally, AEE is improved on average by 56% in terms of estimating the source location. While the existing approaches over or under estimate the number of true sources occasionally, the proposed method shows complete accuracy in this estimation. Our automated results on 26 epileptic patients was in full agreement with the manual diagnosis of the epileptologist.