Automatic geometrical registration of fluorescein retinal angiograms

Abstract

Changes in retinal microvascular perfusion revealed by fluorescein angiography have been reported in patients undergoing coronary artery surgery. Quantification of these changes is important and current techniques depend on careful visual inspection of the angiograms by humans. Computer image processing methods can be used to identify and highlight differences, but geometrical registration of the images is a prerequisite to the comparisons. Automatic methods for locating and matching reference points have therefore been developed. In combination with an iterative process which used least-squares error to calculate the transformation coefficients, subsets of reference points were selected and used to register successfully 20 image pairs.