On Transfer Learning for Classifying COVID-19 in Chest X-Ray Images

Elilson Santos; Lúcio Flavio de Jesus Silva; Omar Andres Carmona Cortes

doi:10.48011/asba.v2i1.991

Elilson Santos Universidade Estadual do Maranhão
Lúcio Flavio de Jesus Silva Universidade Estadual do Maranhão
Omar Andres Carmona Cortes Instituto Federal de Educação, Ciência e Tecnologia do Maranhão

DOI: https://doi.org/10.48011/asba.v2i1.991

Keywords: Deep learning, Transfer learning, Classification, COVID-19

Abstract

COVID-19 is an exceptionally infectious disease caused by severe acute respiratory syndrome. The illness has spread itself worldwide rapidly and can lead to death only in a few days. In this context, investigating fast ways of detection that help physicians in the decision-making process is essential to help in the task of saving lives. This work investigates fourteen convolutional neural network architectures using transfer learning. We used a database composed of 2,928 x-ray images divided into three classes: Normal, COVID-19, and Viral Pneumonia. Results showed that DenseNet169 presented the best results regarding classification reaching a mean accuracy of 94%, a precision of 97.6%, a recall of 95.6%, and an F1-score of 96,1%, approximately.