Assessing Variability of EEG and ECG/HRV Time Series Signals Using a Variety of Non-Linear Methods

Abstract

Time series signals, such as Electroencephalogram (EEG) and Electrocardiogram (ECG) represent the complex dynamic behaviours of biological systems. The analysis of these signals using variety of nonlinear methods is essential for understanding variability within EEG and ECG, which potentially could help unveiling hidden patterns related to underlying physiological mechanisms. EEG is a time varying signal, and electrodes for recording EEG at different positions on the scalp give different time varying signals. There might be correlation between these signals. It is important to know the correlation between EEG signals because it might tell whether or not brain activities from different areas are related. EEG and ECG might be related to each other because both of them are generated from one co-ordinately working body. Investigating this relationship is of interest because it may reveal information about the correlation between EEG and ECG signals. This thesis is about assessing variability of time series data, EEG and ECG, using variety of nonlinear measures. Although other research has looked into the correlation between EEGs using a limited number of electrodes and a limited number of combinations of electrode pairs, no research has investigated the correlation between EEG signals and distance between electrodes. Furthermore, no one has compared the correlation performance for participants with and without medical conditions. In my research, I have filled up these gaps by using a full range of electrodes and all possible combinations of electrode pairs analysed in Time Domain (TD). Cross-Correlation method is calculated on the processed EEG signals for different number unique electrode pairs from each datasets. In order to obtain the distance in centimetres (cm) between electrodes, a measuring tape was used. For most of our participants the head circumference range was 54-58cm, for which a medium-sized I have discovered that the correlation between EEG signals measured through electrodes is linearly dependent on the physical distance (straight-line) distance between them for datasets without medical condition, but not for datasets with medical conditions. Some research has investigated correlation between EEG and Heart Rate Variability (HRV) within limited brain areas and demonstrated the existence of correlation between EEG and HRV. But no research has indicated whether or not the correlation changes with brain area. Although Wavelet Transformations (WT) have been performed on time series data including EEG and HRV signals to extract certain features respectively by other research, so far correlation between WT signals of EEG and HRV has not been analysed. My research covers these gaps by conducting a thorough investigation of all electrodes on the human scalp in Frequency Domain (FD) as well as TD. For the reason of different sample rates of EEG and HRV, two different approaches (named as Method 1 and Method 2) are utilised to segment EEG signals and to calculate Pearson’s Correlation Coefficient for each of the EEG frequencies with each of the HRV frequencies in FD. I have demonstrated that EEG at the front area of the brain has a stronger correlation with HRV than that at the other area in a frequency domain. These findings are independent of both participants and brain hemispheres. Sample Entropy (SE) is used to predict complexity of time series data. Recent research has proposed new calculation methods for SE, aiming to improve the accuracy. To my knowledge, no one has attempted to reduce the computational time of SE calculation. I have developed a new calculation method for time series complexity which could improve computational time significantly in the context of calculating a correlation between EEG and HRV. The results have a parsimonious outcome of SE calculation by exploiting a new method of SE implementation. In addition, it is found that the electrical activity in the frontal lobe of the brain appears to be correlated with the HRV in a time domain. Time series analysis method has been utilised to study complex systems that appear ubiquitous in nature, but limited to certain dynamic systems (e.g. analysing variables affecting stock values). In this thesis, I have also investigated the nature of the dynamic system of HRV. I have disclosed that Embedding Dimension could unveil two variables that determined HRV.