Detecção de lesões vermelhas em imagens de fundo do olho

Abstract

Evaluation of the retina can provide information regarding the presence of ocular diseases such as glaucoma and macular degeneration. Moreover, chronic systematic diseases such as high blood pressure and diabetes can also be detected through of the retina conditions. The red lesions detections in eye fundus images has many relevant applications because they may reveal diseases thet still caused no symptoms. Thus, the presente paper aim to automatically detectd red lesions in eye fundus images. Our proposed method uses mathematical morphology and has two steps, namely: a) coarse detection and b) fine detection of red lesions. The coarse detection step detects dark structures of the retina, such as red lesions, blood vessels and fóvea. On the other hand, the fine detection aims to improvide the detection of red lesions detected in the precious step. The method achieved a mean sensitivity of 75.82% and mean specificity of 93.49% in the fine detection step using a publicy available database. This demonstrates that the method could be useful in a real screening system, and may assist the work of experts in ophthalmology.