|
Abstract:
|
Nowadays, computer simulations can be much faster or less costly than running physical experiments. Therefore computer models can be a stand-alone tool and combined with (typically smaller) data from physical experiments or field observations. In this background, we developed parametric models by decomposing patient-specific longitudinal CT scan images into 3 additive Gaussian models--AAA G&R computer model, inadequacy of computer model and measurement errors. Among all the parameters, calibration parameters assumed to be most essential in the computer model are either unknown or unmeasured in the physical images. Thereupon, the aim of this paper is to calibrate calibration parameters to improve patient-specific predictions. To achieve our aims, we formulate the Bayesian analysis integrating all sources of variation and randomness, train the model to calibrate the patient-specific calibration parameters using patient-specific longitudinal CT scan images, predict the expansion of AAA for each patient in future time and analyze the associated uncertainty in terms of posterior estimation and prediction. Finally, we validate our Bayesian calibration model and show the usefulness.
|
