Serviço web para diagnóstico de retinopatia diabética através de imagens utilizando redes neurais convolucionais

Abstract

Diabetes is one of the fastest growing diseases in recent decades. The number of adults living with this disease has more than tripled in the last 20 years, and estimates indicate that this number will only increase. The biggest problem for people with diabetes are the consequences brought about by high levels of glucose in the blood, ranging from cardiovascular diseases to amputation of limbs. A common consequence in people with diabetes is the development of diabetic retinopathy, a condition that causes swelling and leaks in the blood vessels of the retina, which can even lead to complete loss of vision. Diabetic retinopathy is the main preventable cause of blindness, and this is mainly due to late diagnosis and lack of adequate treatment to reverse the patient’s condition. Based on this scenario, the present work was developed with In order to use Artificial Intelligence to aid in the diagnosis of this pathology, 4 different architectures of Convolutional Neural Networks were tested to perform the diagnosis, namely Lenet-5, Alexnet, Googlenet and Vgg-19, the latter reaching approximately 78% accuracy with sensitivity above 98%. The trained models can be used to assist the diagnosis of this condition and thus making diabetic people with retinopathy undergo early treatment, preventing the development of the disease to its most aggressive form.