MULTIVARIATE TIME-VARYING AUTOREGRESSIVE MODELING OF FETAL SYMPATHO-VAGAL BALANCE THROUGH GESTATION
A processing framework is proposed to model relative changes in fetal sympatho-vagal balance at equally spaced gestational periods. The proposed method is based on a multivariable time-varying autoregression (TVAR) of the beat-to-beat time differences obtained from non-invasive fetal electrocardiographic (ECG) or magnetocardiographic (MCG) measurements. In order to quantify the sympatho-vagal balance at each measured gestational period, the ratio between the standard deviation of normal-to-normal (SDNN) beat intervals and the sum of absolute differences (SAD) is computed. While the SDNN quantifies the overall variability of the sympathetic and vagal systems, the SAD enhances short-term variability components related to vagal control, then the ratio of these two compares with high specificity the overall variability against the short-term vagal component in the time domain. The SDNN/SAD ratio is used to form a new data set by removing short-term variability events, then leaving only those corresponding to longer-term sympatho-vagal balance. The new data set is then analyzed as a dynamical system by fitting it to a suitable multivariate TVAR, and relative changes in the sympatho-vagal balance through the analyzed gestational periods are assumed to be related to the dynamics of the time-varying coefficients of the TVAR. In order to demonstrate the applicability of the proposed method, simulated and real fetal E/MCG data are analyzed. The results show that the modeling approach is able to infer the expected trend seen through sympatho-vagal development.