Normalizing images is good to improve computer-assisted COVID-19 diagnosis

• Autor(a) principal: Santos, Claudio Filipi Gonçalvesdos
• Data de Publicação: 2021
• Outros Autores: Passos, Leandro Aparecido [UNESP], Santana, Marcos Cleisonde [UNESP], Papa, João Paulo [UNESP]
• Tipo de documento: Capítulo de livro
• Idioma: eng
• Título da fonte: Repositório Institucional da UNESP
• Texto Completo: http://dx.doi.org/10.1016/B978-0-12-824536-1.00033-2; http://hdl.handle.net/11449/234306
• topic: Convolutional neural network; Coronavirus; COVID-19

The Coronavirus Disease 2019 (COVID-19) outbreak, caused by the SARS-CoV-2 virus, surprised the whole world in an unprecedented and devastating way, resulting in almost deaths and 2.3 million infections worldwide in less than 4 months. Moreover, the elevate capability of transmission threatens to collapse both the healthy and economic systems from most countries, stressing worse predictions for emerging countries. In such a turbulent scenario, fast diagnosis is essential for a successful treatment and isolation of patients, thus avoiding increasing the number of contaminations. However, traditional methods of detection using polymerase chain reaction are impractical in large scale due to elevate costs, material scarcity, and time demanded for processing. As an alternative, some researchers proposed a machine learning-based diagnosis considering chest X-ray analysis with promising results, thus opening room for possible improvements. This work introduces a different normalization approach that, together with an EfficientNet-B6-inspired neural network, can deal with COVID-19 diagnosis considering chest X-ray images. Experiments provided competitive results considering a lighter and faster architecture, thus fostering research toward COVID-19 detection.