A Multi-objective Evolutionary Algorithm for EEG Inverse Problem

TL;DR

This paper introduces a multi-objective evolutionary algorithm, MOEAAR, for solving the EEG inverse problem without needing empirical parameters, demonstrating improved stability and sparse solutions over classic regularization methods.

Contribution

The paper presents MOEAAR, a novel multi-objective evolutionary algorithm that effectively estimates distributed solutions for EEG inverse problems without empirical parameter tuning.

Findings

MOEAAR outperforms classic methods in localization accuracy.

The evolutionary approach yields relevant sparse solutions.

MOEAAR shows better stability in noisy conditions.

Abstract

In this paper, we proposed a multi-objective approach for the EEG Inverse Problem. This formulation does not need unknown parameters that involve empirical procedures. Due to the combinatorial characteristics of the problem, this alternative included evolutionary strategies to resolve it. The result is a Multi-objective Evolutionary Algorithm based on Anatomical Restrictions (MOEAAR) to estimate distributed solutions. The comparative tests were between this approach and 3 classic methods of regularization: LASSO, Ridge-L and ENET-L. In the experimental phase, regression models were selected to obtain sparse and distributed solutions. The analysis involved simulated data with different signal-to-noise ratio (SNR). The indicators for quality control were Localization Error, Spatial Resolution and Visibility. The MOEAAR evidenced better stability than the classic methods in the reconstruction and localization of the maximum activation. The norm L0 was used to estimate sparse solutions with the evolutionary approach and its results were relevant.