Identifying Pneumonia in SARS-CoV-2 Disease from Images using Deep Learning

Abstract

Deep learning methods are commonly used in various applications now a days. It has also shown its effectiveness in the field of medicine and is actively used as an auxiliary technology. Although the COVID-19 epidemic is an unprepared process in the world, it has been one of the times when alternatives were most needed for diagnosis and treatment. Considering the cost of tests such as PCR in the diagnosis of the COVID-19 epidemic, it has become a need to create economical alternatives. In this context, the main objective of this study was to diagnose and distinguish between COVID-19 and pneumonia cases with CNN-based multiple models using images of chest X-rays. Specifically, three different CNN-based models were used, namely: InceptionV3, ResNet50, and InceptionResNetV2. Moreover, different optimizers were also investigated to identify the best performing one. The models were trained on chest X-ray of 100 patients with COVID-19 and pneumonia. The experimental results showed that the SGD optimizer with the highest accuracy came to the fore with a value of 98.8%.