Metodología de Segmentación de la Estructura Ocular en Imágenes de Fondo de Ojo de Pacientes con Retinopatía Diabética

Abstract

Diabetic retinopathy (RD) is a complication caused by diabetes. This can cause blindness, becoming a worldwide threat due to the affectation of people at an early age and linked to the labor field. The role of medicine in the diagnosis of this disease is associated to the analysis of pathologies (micro aneurysms, hemorrhages and exudates) presented in fundus images. This task is tedious, time-consuming for medical personnel and it makes delays in diagnosis because of the high demand of  RD patients. From the engineering of CAD system designs (Computer-Aided Design) an automated method is proposed for the identification of reference characteristics such as the optical disc, the fovea and the blood vessels. This article presents a methodology for the detection and segmentation of the optical disc and blood vessels. first, a preprocessing stage based on the algorithms of contrast-adaptive histogram equalization (CLAHE) and dynamic diffuse histogram equalization that preserve brightness (BPDFHE) is proposed aiming of improving the fundus image and highlighting the blood vessels for later removal. Next, it is proposed the methodology of segmentation of blood vessels through neural networks that a posteriori, it is going to contribute information for the correct segmentation of the optical disc, which is located through iterative opening and closing morphological operations, demonstrating that it is invariability against pathological lesions of the exudates that coincide in color information with the ocular structure of interest.Once the optic disc is located, the blood vessels inside are removed by inpainting algorithm (Functional Mumford-Shah), which replaces the presence of the blood vessels with statistical information of color discrimination of the optical disc. Finally, the Graph Cut algorithm is implemented for the segmentation of the optical disc.