In brain-computer interfaces (BCI), the detection of different mental states is a key element. In Motor Imagery (MI)-based BCIs, the considered features typically rely on the power spectral density (PSD) of brain signals, but alternative features can be explored looking for better performance. One possibility is the integration of functional connectivity (FC). These features quantify the interactions between different brain areas and they could represent a valuable tool to detect differences between two mental conditions. Here, we investigated the behavior of coherence-based FC features and PSD features, alone and in combination. For a better comparison, we characterized the network centrality of each brain area by computing the weighted node degrees from the estimated FC networks. Our findings show that in both alpha and beta frequency bands, and for almost all the subjects, the fusion of FC network indices and PSD features give better performance. This preliminary results open the way to the use for network-based approaches in BCIs.