Uncertainty quantification and sensitivity analysis for cardiovascular models
MetadataVis full innførsel
The major challenge for applying cardiovascular models in personalized medicine is as follows: How certain are the predictions of a cardiovascular model, taking the natural variations and measurement uncertainty of model inputs into account? To address this challenge, we combined methodologies for uncertainty quantification (UQ) and sensitivity analysis (SA) with cardiovascular models. In a comprehensive guide to UQ and SA, we showed how to quantify the uncertainties of model predictions and the sensitivity of model predictions with respect to model inputs. The approach is exemplified for two clinically relevant models, predicting the following: i.) the severity of coronary artery stenosis, which is a predisposition to stroke and transient ischaemic attack, and ii.) the total arterial compliance, which is considered to be a cardiovascular risk factor, indicating the development of hypertension and atherosclerosis. Furthermore, we developed a framework for UQ and SA in a one-dimensional blood flow model, which can be applied for patient-specific simulations of the cardiovascular system under healthy and diseased conditions. Using this framework, we identified that the aortic arteries play a key role in the development of age-related hypertension. Moreover, we demonstrated how UQ and SA can be applied to guide the selection of the most suitable models, exemplified with the choice of arterial wall models when simulating onedimensional arterial networks. In the future, personalized computer models will be applied in medical practice and become an integral part of cellphone apps for diagnostics, robots for interventions, and clinical software for diagnostics and intervention planning. Consequently, vital decisions will increasingly be based on the predictions of computer models. With this outlook, the reliability of the applied computer models need to be assessed and proven using methodologies such as those presented in this thesis.