Herramienta de ayuda diagnóstica basada en redes neuronales para analizar angiogramas reticulares implementada en el Instituto para Niños Ciegos y Sordos

Abstract

Artificial neural networks are a great tool enabling activities automatically that currently only could be done by humans, this technology leads the field of computer vision and processes of automation. This worked grade seeks to implement a neural network for solving a problem of logistics classification in an institution without profit. In the institution are carried out examinations of reticular angiography on a daily basis and the accumulation of results becomes a tedious job for doctors and the time of waiting for the delivery of results is considerable. So, the design of a tool to help diagnose that allows sorting the reticular angiography based on the probability of being pathological or not pathological; in order to give to doctors in order of most likely to be pathological and try to reduce the speaking time to patients with high probability of pathology. The development of the tool used Python programming language and the library of neural networks Tensorflow and Keras. It was necessary to create a database with images angiographic provided by the institution for the training of the neural network process. Subsequently, it was proposed three alternatives of artificial neural networks that have the capacity to process angiographic information; the next thing was to train and validate the models and compare their performance in not seen data through the construction of a confusion matrix. Finally, it was chosen the model as a best result in the confusion matrix and realized the deployment of the model by means of a GUI which connects to the tool with the user