Non-parametric Source Reconstruction via Kernel Temporal Enhancement for EEG Data

Source reconstruction from EEG data is a well know problem in the neuroscience field and affine areas. There are a variety of applications that could be derived form an adequate source reconstruction of the cerebral activity. In recent years, non-parametric methods have been proposed in order to improve the reconstruction results obtained from the original Low Resolution Tomography (LORETA) like approaches. Nevertheless, there is room for improvement since EEG data could be processed to enhance the reconstruction process via some temporal and spatial transformations. In this work we propose the use of a Kernel-based temporal enhancement (kTE) of the EEG data for a preprocessing stage that improves the results of source reconstruction into the non-parametric framework. Three metrics of source error localization named as Dipole Localization Error (DLE), Euclidean Distance (ED) and Dipole dispersion (DD) are computed for comparing the performance of swLORETA in different scenarios. Results shows an evident improvement in the reconstruction of brain source from the proposed kTE in comparison to the state of art non-parametric approaches.