Assessment of reduced‐order unscented Kalman filter for parameter identification in 1‐dimensional blood flow models using experimental data

• Published in: International journal for numerical methods in biomedical engineering 2017-08, Vol.33 (8), p.e2843-n/a
• Main Authors: Caiazzo, A., Caforio, Federica, Montecinos, Gino, Muller, Lucas O., Blanco, Pablo J., Toro, Eluterio F.
• Format: Article
• Language: English
• Subjects: 1‐dimensional model; Algorithms; Aorta; Arteries - physiology; Biomedical Engineering - methods; Blood flow; Computer Simulation; Elastic Modulus; Experimental data; finite volume method; Flow measurement; Hemodynamics - physiology; Humans; In vitro methods and tests; Kalman filter; Mathematical models; Mechanical properties; Models, Cardiovascular; Modulus of elasticity; Nonlinear Dynamics; Parameter estimation; Parameter identification; Pressure; Reduced order filters; Reduced order models; Reproducibility of Results; Sensitivity analysis; Vascular Stiffness; Wall thickness
• ISSN: 2040-7939; 2040-7947
• DOI: 10.1002/cnm.2843

This work presents a detailed investigation of a parameter estimation approach on the basis of the reduced‐order unscented Kalman filter (ROUKF) in the context of 1‐dimensional blood flow models. In particular, the main aims of this study are (1) to investigate the effects of using real measurements versus synthetic data for the estimation procedure (i.e., numerical results of the same in silico model, perturbed with noise) and (2) to identify potential difficulties and limitations of the approach in clinically realistic applications to assess the applicability of the filter to such setups. For these purposes, the present numerical study is based on a recently published in vitro model of the arterial network, for which experimental flow and pressure measurements are available at few selected locations. To mimic clinically relevant situations, we focus on the estimation of terminal resistances and arterial wall parameters related to vessel mechanics (Young's modulus and wall thickness) using few experimental observations (at most a single pressure or flow measurement per vessel). In all cases, we first perform a theoretical identifiability analysis on the basis of the generalized sensitivity function, comparing then the results owith the ROUKF, using either synthetic or experimental data, to results obtained using reference parameters and to available measurements. This work considers a parameter estimation approach on the basis of the reduced‐order unscented Kalman filter in the context of one‐dimensional blood flow models, investigating the effects of using real measurements versus synthetic data for the estimation procedure. The filter is assessed considering the results of an in vitro model of the human arterial network and the available experimental measurements, comparing the estimation results with an identifiability analysis on the basis of the generalized sensitivity function and considering flow and pressure observations.