Performance Comparison Between GPDC And PCMI For Measuring Directionality Of Neural Information Flow

ObjectiveWhen brain is studying or thinking, different brain areas are connected each other by means of neural oscillations. In order to measure the patterns of neural oscillations, the analysis of neural information flow (NIF) has been developed. It can reflect real-time changes in brain connection.In recent years, there have been many algorithms, which are used to evaluate the strength and direction of neural information flow, so as to be able to determine the connection status in different brain regions when brain at work. However, the measurements used by different approaches may generate diverse results and even conflicting results. Therefore, by comparing with different methods, the determination of practice and the conclusion can be a certain guiding. In this study, two approaches of NIF measurement have been compared each other. One is a linear algorithm, general partial directional coherence (gPDC), which is a representative approach of the method. Another is a nonlinear algorithm, permutation conditional mutual information, which is a representative of the nonlinear method. In order to compare the characteristics of these two kinds of approaches in the estimation of the strength and directionality of neural information flow and provide a number of guidance for real data analysis, a mathematical model has been employed to simulate real EEG data.MethodsNeural mass model (NMM) was introduced to simulate real EEG data. Several comparisons were performed, which includes the sensitivity of the algorithms, the error of measurement and the influence of the length of data points. Finally, we analyzed real data (LFPs) to support our conclusions, which were drew by numerical experiment.Results1. The sensitivity of PCMI is better than that of gPDC. Compared with gPDC, PCMI gets more reliable results in the state of weaker coupling strength. Consequently, PCMI can be used to measure the coupling strength when the interaction is weaker.2. The sensitivity of gPDC is a binary function, which is influenced by both directions. However, the sensitivity of PCMI is a unary function, which is only determined by the direction measured by PCMI. Thus, PCMI is more suitable to measure the coupling strength.3. In unidirectional model that is under the one-way driving conditions the gPDC is better than the PCMI in estimating the coupling strength. At the same time, the gPDC is better than the PCMI in reflecting the changes in terms of coupling strength.4. In bidirectional model that is under the mutual driving conditions the PCMI is better than the gPDC in estimating the coupling strength. Meanwhile, the gPDC better reflects the changes in connection strength5. Compared with gPDC, PCMI needs shorter length of data points. In order to obtain stable results, gPDC needs longer length of data points6. The results of real-data analysis show that gPDC is more suitable to detect the changes of connection strengthConclusions1. The gPDC is more suitable to detect the changes of NIF connection strength;2. The PCMI is more suitable to estimate the connection strength of NIF.