Desarrollo de un aplicativo web utilizando redes neuronales convoluciones para el reconocimiento de enfermedades pulmonares mediante el análisis de radiografías.

Abstract

In recent years, the advancement of AI has achieved that we all have the ability to make and train different convolutional neural networks that are capable of making their own decisions, this project aims to develop a web application for the recognition of lung diseases such as COVID-19. 19, Tuberculosis and Pneumonia using convolutional networks through the analysis of radiographs in order to obtain more efficient results in the shortest possible time, since today the time of analysis of radiographs is too long at the time of obtaining results. To elaborate the convolutional neural network, a database was created with a total of 4000 images or radiographs and they were classified into four groups and each group contains a total of 1000 radiographs: the first group of disease radiographs is Covid, the second group of X-rays of diseases is Pneumonia, the third group of X-rays of diseases is Tuberculosis and the fourth group of X-rays is Normal. For the elaboration of the image database, X-rays were obtained from the Kaggle page, which is a free platform where there are countless data sets of any subject, and the rest of the X-rays were collected from several reliable pages. from Internet. Currently there are various methods of recognizing objects or diseases, for this thesis the tools used for the elaboration and preparation of the neural network were Google Colab, the Keras library and Tensorflow. Subsequently, several tests were carried out with the previously elaborated and prepared network. Two convolutional networks were prepared with various indicators such as the number of epochs, which in this case were 100 and 200, and also the name of the optimizers, one was Adam and the other was SGD, in order to analyze who provides a better training Once the training of the convolutional network was done, multiple tests were carried out within Google Colab, and various metrics were also applied, such as the confusion matrix, in order to select the most optimal convolutional network.