Independent component analysis of sparse-transformed EEG signals for ADHD/normal adults´ classification

The present study investigates the EEG source differences between adults with ADHD and aged match controls. The processing method is based on sparse representation of electrode signals and complex-valued independent component analysis with a robust measure of sparseness. Combination of scalp topography, estimated dipole source location and spectral patterns of resulted ICs were used to k-means clustering and identification of near-equivalent ICs across subjects. Several frequency features were extracted from clustered ICs and individually submitted to k-nn classifier. The best resulted accuracy was 86.36% using f_mean feature at R-parietal cluster. Eight pairs of features resulted in such accuracy. The method used in this study not only improves the participant´s classification accuracy compared to reference analysis, but also better identifies the dynamic of brain source signals than time-domain ICA algorithms.