The analysis parameters for the characterization of heart rate variability (HRV) within a very short time (＜1 min) usually exhibit complicate variation patterns over time, which may easily interfere the judgment to the status of the cardiovascular system. In this study, long-term HRV sequence of 41 cases of healthy people (control group) and 25 cases of congestive heart failure (CHF) patients (experimental group) was divided into multiple segments of very short time series. The variation coefficient of the same HRV parameter under multiple segments of very short time series and the testing proportion with statistically significant differences under multiple interclass t-test were calculated. On this account, part of HRV analysis parameters under very short time were discussed to reveal the stability of difference of the cardiovascular system function under different status. Furthermore, with analyzing the receiver operating characteristic (ROC) curve and modeling the artificial neural network (ANN), the classification effects of these parameters between the control group and the experimental group were assessed. The results demonstrated that ① the indices of entropy of degree distribution based on the complex network analysis had a lowest variation coefficient and was sensitive to the pathological status (in 79.75% cases, there has statistically significant differences between the control group and experimental group), which can be served as an auxiliary index for clinical doctor to diagnose for CHF patient; ② after conducting ellipse fitting to Poincare plot, in 98.5% cases, there had statistically significant differences for the ratio of ellipse short-long axis (SDratio) between the control group and the experimental group; when modeling the ANN and solely adopting SDratio, the classification accuracy to the control group and experimental group was 71.87%, which demonstrated that SDratio might be acted as the intelligent diagnosis index for CHF patients; ③ however, more sensitive and robust indices were still needed to find out for the very-short HRV analysis and for the diagnosis of CHF patients as well.
The study of complex networks has become a hot research area of electroencephalogram signal. Electroencephalogram time series generated by the network keeps node information of network, so studying the time series from the network can also achieve the purpose of study epileptic electroencephalogram. In this paper, we propose a method to analyze epileptic electroencephalogram based on time series which is based on improved k-nearest neighbor network. The results of the experiment showed that studying power spectrum of time series from network was easier than power spectrum of time series directly generated from the original brain data to distinguish between normal controls and epileptic patients. In addition, studying the clustering coefficient of improved k-nearest neighbor network was able to distinguish between normal persons and patients with epilepsy. This study can provide important reference for the study of epilepsy and clinical diagnosis.