Deep Learning for the Detection of Covid-19 Through Test Image Analysis Fast (#1138)

Authors

Patiño-Pérez, Darwin

Jácome-Morales, Gladys

Cedeño-Rodríguez, Juan

Botto-Tobar, Miguel

Icaza-Rivera, Dalva

Burgos-Robalino, Freddy

Sarmiento-Barreiro, Liliana

Abstract

Currently, rapid tests for the detection of COVID-19 are a very useful mechanism and used as an option for the rapid detection of an ongoing infection by COVID 19, since other diagnostic methods are more complex, and require analysis, time and resources. In the health area, it has been decided to use artificial intelligence tools that help detect symptomatic and asymptomatic patients, to prevent future spread. Convolutional neural networks (CNN) are the most convenient since they can learn by themselves to identify features within images. However, it is necessary to train a convolutional neural network using image datasets. Since there is a small dataset of just two thousand images, a technique was used to randomize and rescale the images to greatly increase the dataset and use it as training data in a CNN, thus enhancing its performance. By obtaining the generated images, a CNN model was created, which was carried out in the Google Colab environment, using the python programming language and machine learning libraries such as Keras, Tensorflow and OpenCV, through automatic learning the model learned to predict by classification, obtaining an accuracy of 97% in prediction with a loss of 6.69% and without having fallen into overtraining, its use was recommended since its level of generalization reached 84%.