Open AccessAutomatic Detection of COVID-19 in X-Ray Images Using Fully-Connected Neural Networks
Author(s) -
Élisson De Carvalho,
Raian Malta,
Alessandra Martins Coelho,
Matheus de Freitas Oliveira Baffa
Publication year - 2020
Language(s) - English
Resource type - Conference proceedings
DOI - 10.5753/wvc.2020.13478
Subject(s) - computer science , visualization , covid-19 , artificial intelligence , artificial neural network , computer aided diagnosis , adjacency list , medical imaging , pattern recognition (psychology) , computer vision , image segmentation , image (mathematics) , disease , medicine , algorithm , pathology , infectious disease (medical specialty)
The coronavirus pandemic remains a problem of worldwide interest. The diagnosis of COVID-19 is difficult due to its high rate of occurrence and the limited number of test kits. Medical imaging is already widespread and has been used to quickly provide lung visualization. It’s needed some expertise from the radiologist to detect elements in the image that allow differentiating the sick and healthy patterns. Therefore, our goal with this paper is to provide a computer-aided diagnosis tool to help radiologists to accurately diagnose the COVID-19 using XRay images. For that, a model based on Fully-Connected Neural Networks was proposed for the detection of patients infected with coronavirus, through the analysis of texture characteristics, such as Haralick and Threshold Adjacency Statistics (TAS) descriptors, extracted from chest X-Ray images. Using 10-Fold Cross-Validation, the proposed method achieved an accuracy of 98.39%, showing itself as an option to aid the disease diagnosis.