Functional brain network based on improved ensemble empirical mode decomposition of EEG for anxiety analysis and detection

Most anxiety studies are based on the features of isolated electroencephalography (EEG) electrode, ignoring that the essence of EEG is the signals overlap from different neurons. Therefore, it's difficult to find abnormal topological alters of the brain from independent neuron. To this end, this paper proposes an anxitey analysis and detection framework of brain function network (BFN) based on improved Ensemble empirical mode decomposition (EEMD). Additional adaptive white noise is adopted to solve the pattern aliasing problem, and several intrinsic mode functions (IMF) for simulating independent neuron signals are obtained. Binary BFN is constructed on different IMF based on phase lag index (PLI) and proportional threshold strategy. Then complex network method is used to analyze the topology alters and attributes of BFN, and explore the potential biomarkers of anxiety detection. In addition, to evaluate the effectiveness of these potential biomarkers for anxiety detection, support vector machine (SVM) classifier is used as an evaluation tool and highest detection accuracy of 92.38% was obtained. Meanwhile, the analysis results show that the adaptive amplitude white noise can affect all extreme points, and improved EEMD can more effectively decompose EEG signals. Functional synchrony altered significantly in the frontal, temporal and central region of the left brain. Our research findings also show that the BFN of anxiety patients shows a tendency of randomization.