Digital curvature estimation for left ventricular shape analysis

摘要

Curvature estimation from noisy digital curves is addressed for shape analysis of angiographic left ventricle (LV) images. A simple and fast algorithm, based on Fourier series approximation, changes the harmonic number and filter window along the closed (or partially open) contours to optimize smoothness and reconstruction errors. The curvatures estimated on a set of ventricle-shaped models are significantly closer to their analytical values than those assessed by four other methods. Preliminary results on LV data are presented.