A comparison of effective connectivity methods using different performance metrics

Different analysis methods have been developed to determine brain connectivity patterns. To select suitable methods depending on application contexts, it is essential to evaluate different methods using suitable performance metrics. We propose three application-oriented metrics which enable to measure multivariate causality qualitatively. Using the proposed metrics, the most used analysis methods (Directed Transfer Function, Partial Directed Coherence, Granger-Geweke Causality) are compared on synthetic electroencephalographic data with a predefined causality structure. Furthermore, the performances obtained by using all metrics are evaluated. Such analysis allows us to select the most stable analysis method. Also, the optimal metric can be found by comparing the performances obtained by using different metrics with the graph-based performance in the graphically displayed predicted network.