Artificial Intelligence, Case Study: Detection of Diabetic Retinopathy Through a Neuronal Networks in Citizens of Bogotá-Colombia

Abstract

This document shows a brief history and a simple definition of artificial intelligence, to observe the concept of artificial neural networks in order to observe the case of diabetic retinopathy in Colombia, to understand the factors that facilitate this disease, it will also be studied the classification of healthy people and patients with the condition through a supervised artificial neural network that was explored and studied with different architectures in which the hidden layers, the training algorithm, the transfer function and the number of neurons were varied, where it was observed that the Levenberg-Marquadt method was the most suitable for the classification of affected patients, due to the speed of the algorithm in the training and validation of the data, which had a more agile response time than the algorithms with optimization of gradient descent, likewise it was visualized that by having a reduced number of neurons in l. The hidden part layers help to reduce the error, reduce the computational load of the system and when complemented with a suitable objective function such as logsig, it allows an approximation to the desired answer. For the diabetic retinopathy problem, it was observed that the best architecture is ARC [30 1], which has 31 neurons, since it has the lowest validation error of 24%.