Automatic detection of microaneurysms in diabetic fluorescein angiography

Abstract

A computerized method for the detection of microaneurysms (MA) in fluorescein angiograms is proposed, using the concepts of mathematical morphology. The MA which are almost circular particles, are extracted from the image by performing different "top-hat transformations". Some particles, however, may then be detected inside the nonhomogeneous vessels, and it is necessary to be able to extract the vasculary net. The MA which present fuzzy boundaries (due to the leakage of fluorescein) are the most difficult to extract. The algorithm has been tested on 25 angiograms, with 1 045 MA analysed. A comparison between the automatic counting procedure and three manual methods of counting has been made to prove the robustness of the proposed method.