Robust estimation of sparse EEG source based on Laplacian distribution

Locating the neural electric current sources and calculating their magnitudes through scalp electroencephalography (EEG) offer a way to explore the transient brain activity inside the head. However, because the EEG recordings are inevitably influenced by the outliers caused by eye blinking artifacts and head movements, the inferred cortical source activity are usually biased, which may limit its applications in both neural science study and clinical analysis. In our current study, by assuming that the model noises of EEG inverse problem follow the Laplacian distribution and searching the optimal model coefficients in the sparse space, an outlier robust sparse source location method was proposed. To verify its robustness to complex noises and its effectiveness in capturing the activations of multiple cortical sources, we conducted a simulation experiment. By measuring the differences between the predefined cortical sources and the estimated ones through four indexes, we quantitatively assessed its performance. The experimental results consistently proved that comparing with commonly used source estimation technologies such as wMNE, sLORETA, FOCUSS, and LASSO, our proposed method can resist the influence of outliers efficiently and recovered the EEG sources closer to those predefined. These observations verified its powerful capability of offering an alternative for EEG source imaging, especially in the condition that the spiking noise are frequently occurred.