Application of dense and convolutional neural networks for COVID_19 detection in Xray images

Abstract

Convolutional neural networks (CNNs) have great potential in solving classification problems with images. The present research aims to present reduced models that allow identifying cases of pneumonia and COVID-19 in chest X-ray images (anterior-posterior), offering a broad perspective of the interest of tools that provide medical support and assistance. The capacity and size of the models were reduced until obtaining a perfect option to be deployed locally in devices with limited resources. The proposed algorithms were developed in Google Colab using the Python programming language, applying dense and convolutional neural networks to different layers until obtaining a low error rate, to later diagnose if the patient has COVID-19. To do this, a set of 603 high-resolution images from public databases (see in https://www.cell.com/cell/fulltext/S0092- 8674(18)30154-5 and https://github.com/ieee8023/covid-chestxray-dataset) is used, divided into 403 images for training, 200 images for testing and 12 images for validation. The tool designed with a convolutional neural network of 13 layers proposes the integration of machine learning (Machine Learning) as a support in the medical diagnosis process, with an accuracy of 94.73% can become a tool that provides greater speed when giving a diagnosis.