Parameter estimation of flow-measurement in digital angiography

Abstract

The purpose of angiographic procedures used in cardiovascular interventions is to classify the patient's potential of regeneration after strokes caused by dead blood cells in the main arteria. The flow of blood into heart's capillaries is measured using x-ray radiometry with contrastive fluids. One quick and reliable method for estimating this potential could save lives and would allow further treatments to be more accurately planned. Our task was to fit a 5-parameter Gamma function to the intensity samples extracted from the x-ray angiograms. The estimation of this function's parameters is hard given that the raw data set is heavily polluted with several different types of noise. Our complete solution has four main parts which have also been successfully verified and validated. First, we propose a solution for eliminating the noise by applying a specially designed moving window Gauss filter. Secondly, we have designed an algorithm for computing a good initial guess for the Levenberg-Marquardt optimizer in order to achieve the required precision. Third, an algorithm is proposed for selecting significant points on the smoothed data set with an interval-based classification method. Finally, we apply the LM algorithm to compute the solutions in a nonlinear least squares way. We have also formulated an algorithm based on interval arithmetic which can be effectively used for comparing nonlinear least-squares fit results and assign goodness values based on their residuals. This method has been used for measuring improvements during the development. We must emphasize that the proposed algorithms are distinct, they can be used in other applications together or separately since they are generally applicable, they do not depend on specialties of the presented application.