Uma arquitetura digital baseado em aprendizagem de máquina para diagnóstico de doenças oculares: um estudo aplicado ao glaucoma

Abstract

The development of new technologies has been essential for the diagnosis and treatment of eye diseases. In this context, using machine learning (AM) and deep learning (DL) techniques for the classification and detection of ocular analogies has resulted in recent research. In this perspective, the present’s objective is to develop a technical classification service for DL, web technologies, and cloud services, making a case study on glaucoma. Glaucoma is an asymptomatic disease that can cause irreversible blindness if diagnosed late. The type of image used in the analysis was retinography since its acquisition is inexpensive, and a non-dependent operator performs the examination. That is, it does not require the presence of the doctor. As a result, an intelligence architecture was designed that implemented and trained several DL architectures. The learning transfer technique (transfer learning) was used to reduce training time and optimize the method. Another essential strategy is the data augmentation process (data argumentation) to minimize network overfitting, thus preventing the network from generalizing in the test set. A cloud service was developed, a platform that uses pre-trained models to identify new images’ according to the input’s mass of data.