Development of a Registration Technique with Nonlinear Geometric Warping in DSA Images

Abstract

Misregistration artifact is the major cause of image degradation in digital subtraction angiography(DSA). As three-dimensional movement of the body causes uneven misregistration on two-dimensional images, conventional linear correction is not sufficient. We developed a nonlinear geometric warping technique. Using a local least-square matching method, we obtained the shift values sampled at 49 reference points in a 512×512 matrix. They were then fitted to the spline function to calculate pixel-by-pixel shift values. With these values, nonlinear geometric warping of the mask image followed. As a result, misregistration artifact was almost non-existent in the processed images, and vascular anatomy was clearly delineated even in the skull base region. This newly proposed method is a simple, practical way to reduce misregistration artifacts in DSA.