Using Kalman Filtering to Predict Time-Varying Parameters in a Model Predicting Baroreflex Regulation During Head-Up Tilt

作者：Matzuka Brett; Mehlsen Jesper; Hien Tran; Olufsen Mette Sofie^*

来源：IEEE Transactions on Biomedical Engineering, 2015, 62(8): 1992-2000.

DOI：10.1109/TBME.2015.2409211

• Baroreflex regulation
• cardiovascular
• DRAM
• ensemble transform Kalman filter
• head-up tilt
• Kalman filtering
• parameter estimation

The cardiovascular control system is continuously engaged to maintain homeostasis, but it is known to fail in a large cohort of patients suffering from orthostatic intolerance. Numerous clinical studies have been put forward to understand how the system fails, yet noninvasive clinical data are sparse, typical studies only include measurements of heart rate and blood pressure, as a result it is difficult to determine what mechanisms that are impaired. It is known, that blood pressure regulation is mediated by changes in heart rate, vascular resistance, cardiac contractility, and a number of other factors. Given that numerous factors contribute to changing these quantities, it is difficult to devise a physiological model describing how they change in time. One way is to build a model that allows these controlled quantities to change and to compare dynamics between subject groups. To do so, it requires more knowledge of how these quantities change for healthy subjects. This study compares two methods predicting time-varying changes in cardiac contractility and vascular resistance during head-up tilt. Similar to the study by Williams et al. [51], the first method uses piecewise linear splines, while the second uses the ensemble transform Kalman filter (ETKF) [1], [11], [12], [33]. In addition, we show that the delayed rejection adaptive Metropolis (DRAM) algorithm can be used for predicting parameter uncertainties within the spline methodology, which is compared with the variability obtained with the ETKF. While the spline method is easier to set up, this study shows that the ETKF has a significantly shorter computational time. Moreover, while uncertainty of predictions can be augmented to spline predictions using DRAM, these are readily available with the ETKF.

• 出版日期2015-8