Research On Brain Signals Based On Entropy

Nonlinear dynamics are playing a more significant role in brain signal research, and the entropy methodology has broad application prospects and practical value. In recent years, researchers have proposed a variety of entropy theories, which have a broad range of applications in various signal analysis. To explore entropy algorithm application on brain signal, this paper combines with other theories to improve entropy algorithms, referring to brain signal characteristics.This paper begins with the background knowledge about brain signals, and then summaries some popular entropy algorithms used in brain research, including Shannon entropy, approximate entropy, sample entropy and transfer entropy. This paper also provides a feasible method of calculating transfer entropy. Based on the above theories, the following researches were carried out in this paper:First, a kind of epilepsy classification method based on the multi-scale sample entropy, incorporating t-test and support vector machine(SVM) was designed. The method was applied to distinguish epileptic attack electroencephalography(ECoG) with interictal one. The accuracy can reach 98%. After that, analyze the multi-scale sample entropy of schizophrenic patients and the normal magnetoencephalogram(MEG), and plot the brain topographic map. Make the conclusion that the sample entropies of the patients near the temporal lobe are larger than the ones of the normal, but less than in parietal area.Second, apply the relative transfer entropy on the EEG of the alcoholics. Simulations show that the relative transfer entropy of the alcoholics is less than that of the normal. Therefore, as an irreversible characteristic parameter of physical process, the relative transfer entropy has a good result on distinguishing the alcoholics EEG. Meanwhile, discuss how sensitive the algorithm is to the sample length and noise. The results illustrates the relative transfer entropy has a strong robustness to sample length, and has a good anti-noise property.Third, construct a brain function network based on cross approximate entropy which is used to analyze the normal and the patients with epilepsy. The results show that subjects and patients with epilepsy all can constructed a good network under different rhythm. Further explore some complex network measurements of the normal and patients and provide the statistical figure of small-world characteristics, network density, network efficiency and overall positive and negative match degree, and compare these measures in detail.Fourth, based on the transfer entropy and complex network theory, a brain transfer entropy effective connective network is designed to describe information transfer and coupling strength among brain regions. The simulation proves the transfer entropy can exactly estimate the nonlinear coupling strength and direction of information transfer between the two signals, also proves the transfer of entropy has a good anti-noise characteristics. Then, a construction method of brain effective connective network based on transfer entropy is presented. This method is applied to analyze the EEG of epilepsy patients and normal. The results showed that brain effective connective networks based on transfer entropy may be established for epilepsy patients and the normal. Indeed, there are differences between them. Transfer entropy effective connective network may become an effective new method in epilepsy pathogenesis research.