Ground glass opacity detection convolutional neural network method efficient net-b0 architecture in corona virus disease 2019 cases

Abstract

COVID-19 is an infectious disease caused by a new type of coronavirus. The gold standard diagnosis of COVID-19 is a PCR test. Thorax radiographic examination is an alternative to the diagnosis of COVID-19 against initial screening. Thorax radiography sensitivity is lacking in GGO detection of COVID-19. CNN's deep learning architecture EficientNet-B0 has the ability to represent the same diagnostic prediction features as radiologists. The purpose of this study resulted in CNN's EfficientNet-B0 architecture that is able to classify radiographic images of the thorax automatically in detecting GGO COVID- 19 and has similar results with the results of the doctor's experiment. This type of research is a quasi-experimental design Post-test Only Control Group Design. Build the CNN EfficientNet-B0 architecture through the MATLAB R2021a program. The sample totaled 78 radiographic images of the thorax. The data analysis is a statistical test of Chi-Square. Research proves CNN's performance of the EfficientNet-B0 architecture is excellent in detecting GGO COVID-19 with an AUC value of 0.994 range (0.90-1.00). The deep learning model can be considered as an alternative tool for establishing a diagnosis of COVID-19 with the condition that the verification of the diagnosis decision remains based on the confirmation and justification of the radiologist.