ABSTRACT
The Sars-Cov2 virus has caused the worst health emergency of the last decade. Furthermore, new strains make the fight against COVID-19 appear far from over. The virus causes a severe acute respiratory syndrome that can lead to death. Effective identification of lung damage by chest radiography using deep learning methods could be advantageous for imaging physicians in differentiating people who need to be admitted to an intensive care unit (ICU) from people that don’t require medical attention, to avoid the collapse of health systems. This article describes the development of a deep learning model to classify and assess lung injuries with a protocol for lung injury quantification. The model is based on U-Net segmentation and injury classification according to the RALE score system. Kaggle platform was used to obtain the chest radiography dataset and MATLAB to generate the mask dataset for training. Finally, each lung is divided in 4 quadrants for lesion quantification. An accuracy of 92.86% was obtained in the segmentation process and 100% in the process of classifying levels of lung lesions. © 2022, Springer Nature Switzerland AG.