A Dual State-Parameter Ensemble Kalman Filter for Patient-Specific Cardiovascular Modeling

Developing models for patient-specific healthcare usually requires extensive parameter tuning to produce data that adequately matches an individual's clinical measurements. This tuning process is frequently non-trivial: the parameter set is usually large, and the parameters often have nonlinear relationships with both model predictions and each other. Adjusting the parameters in an ad-hoc way is therefore quite costly and requires significant end-user expertise. To avoid this hurdle, this work develops and discusses an ensemble Kalman filter (EnKF) for automated parameter estimation in a reduced-order cardiovascular model. Specifically, we construct a dual EnKF to independently filter the state and parameter vectors. The performance of the estimator is demonstrated using measurements from both healthy and pathological patients, and attention is given to methods for improving its robustness.