Unique Parameter Identification for Model-Based Cardiac Diagnosis in Critical Care
Lumped parameter approaches for modeling the cardiovascular system typically have many parameters of which many are not identifiable. The conventional approach is to only identify a small subset of parameters to match measured data, and to set the remaining parameters at population values. These values are often based on animal data or the “average human” response. The problem, is that setting many parameters at nominal fixed values, may introduce dynamics that are not present in a specific patient. As parameter numbers and model complexity increase, more clinical data is required for validation and the model limitations are harder to quantify. This paper considers the modeling and the parameter identification simultaneously, and creates models that are one to one with the measurements. That is, every input parameter into the model is uniquely optimized to capture the clinical data and no parameters are set at population values. The result is a geometrical characterization of a previously developed six chamber heart model, and a completely patient specific approach to cardiac diagnosis in critical care. In addition, simplified sub-structures of the six chamber model are created to provide very fast and accurate parameter identification from arbitrary starting points and with no prior knowledge on the parameters. Furthermore, by utilizing continuous information from the arterial/pulmonary pressure waveforms and the end-diastolic time, it is shown that only the stroke volumes of the ventricles are required for adequate cardiac diagnosis. This reduced data set is more practical for an intensive care unit as the maximum and minimum volumes are no longer needed, which was a requirement in prior work. The simplified models can also act as a bridge to identifying more sophisticated cardiac models, by providing a generating set of waveforms that the complex models can match to. Most importantly, this approach does not have any predefined assumptions on patient dynamics other than the basic model structure, and is thus suitable for improving cardiovascular management in critical care by optimizing therapy for individual patients.