Dipartimento di Ingegneria informatica, automatica e gestionale

Group level characterization of brain networks still represents an open issue in modern neuroscience. Investigating the functional mechanisms underlying the complexity of the human brain requires an analytical way to efficiently integrate information from multiple subjects, while properly handling the intrinsic inter-subject variability. Here we investigated the potentiality of the PARAllel FACtorization (PARAFAC) algorithm for the extraction of grand average brain connectivity matrices from simulated EEG datasets. Synthetic data were parametrized according to different levels of three parameters: network dimension (NODES), number of observations (SAMP-SIZE) and noise (SWAP-CON) in order to investigate the way they affect the reconstruction of grand average networks. Robustness to noise as well as the proper modulation of informative content have here been proved empirically, revealing that the best performances in terms of FPR, FNR and AUC were achieved for great number of observations and low noise level.