Diagnóstico de retinopatía diabética en imágenes digitales mediante máquinas de vectores de soporte

Abstract

The diagnosis of diabetic retinopathy is a complex problem, addressed by both medicine and computer science, through techniques such as digital image processing and machine learning algorithms. Traditional diagnostic methods suggest the acquisition of fundus images of patients, using invasive procedures to highlight the features of the retina. In this regard, specialists identify three patterns that serve as a basis for studies on diabetic retinopathy: red blood cell deformation, the presence of exudates, and aneurysms. However, the difficulty in interpreting image information has posed a challenge for medical specialists and has hindered the development of reliable computational tools. This work presents an initial approach to addressing the problem, based on feature extraction from retinal images of both healthy individuals and patients with diabetic retinopathy, which are processed by a Support Vector Machine classifier with the aim of producing a tool capable of distinguishing between healthy patients and those with the condition. Experimental studies show an efficiency of 56% in diagnosis and 54% in the (a priori) detection of the condition. This represents an opportunity for improving the classifier, as well as the openness toward the implementation of alternative computational methods, in order to ensure high reliability of the tool for its application in the healthcare sector.