M/EEG source imaging via the spatiotemporal Kalman filter and its applications in epileptology

Successful epilepsy surgery requires the correct identification of the epileptogenic zone in the brain. Epilepsy generators in the brain may be localized via source reconstruction from surface electroencephalography. This requires the solution of the EEG inverse problem. The EEG inverse problem can be reformulated as a state estimation problem, which is then solved via Kalman filtering. The brain dynamics and the measurement process are modeled via appropriate state-space models. In this work, the spatiotemporal Kalman filter, a modified Kalman filter to tackle the high-dimensional EEG inverse problem, is further developed and applied to solve the EEG inverse problem in epileptology. In this work, the roles of the brain discretization and the Laplacian operator in the improvement of the accuracy of the Kalman filter are investigated. Additionally, the Kalman filter is applied for the source reconstruction of epileptiform discharges and the onset of focal seizures. The performance of the Kalman filter is evaluated for the case of low-density EEG recordings. For the case of high-density EEG, a dimensionality reduction approach is implemented for the Kalman filter to improve its localization accuracy and computational speed. In addition to that, we introduced a model for multimodal fusion of simultaneously-measured magneto- and electroencephalography within the framework of source reconstruction. Finally, the localization of activity from subcortical brain structures in achieved via a regional form of the spatiotemporal Kalman filter. Compared to standard methods, the dynamical solution of the EEG inverse problem via Kalman filtering offers potential improvements in the accuracy and spatial resolution of source reconstruction of brain sources from surface recordings.