Evaluation of Brain Source Localization Methods Based on Test-Retest Reliability With Multiple Session EEG Data

Various EEG source localization methods have been proposed for functional brain research. The evaluation and comparison of these methods are usually based on simulated data but not real EEG data, as the ground truth of source localization is unknown. In this study, we aim to evaluate source localization methods quantitatively under the real situation.

We examined the test-retest reliability of the source signals reconstructed from a public six-session EEG data of 16 subjects performing face recognition tasks by five mainstream methods, including weighted minimum norm estimation (WMN), dynamical Statistical Parametric Mapping (dSPM), Standardized LOw Resolution brain Electromagnetic TomogrAphy (sLORETA), dipole modeling and linearly constrained minimum variance (LCMV) beamformers. All methods were evaluated in terms of peak localization reliability and amplitude reliability of source signals.

In the two brain regions responsible for static face recognition, all methods have promising peak localization reliability, with WMN showing the smallest peak dipole distance between session pairs. The spatial stability of source localization in the familiar face condition is better than those in the unfamiliar face and the scrambled face conditions in the face recognition areas in the right hemisphere. In addition, the test-retest reliability of source amplitude by all methods is good to excellent under the familiar face condition.

Stable and reliable results for source localization can be obtained in the presence of evident EEG effects. Due to different levels of a priori knowledge, different source localization methods have different applicable scenarios.

These findings provide new evidence for the validity of source localization analysis and a new perspective for the evaluation of source localization methods on real EEG data.