Detection of COVID-19 From A New Dataset Using MobileNetV2 and ResNet101V2 Architectures

© 2022 IEEE.The aim of this study is to measure to what extent deep learning architectures are successful in classification by using a new data set consisting of lung CT data collected from ill/healthy individuals in order to diagnose COVID-19. To achieve this purpose, the researcher prepared a new data set involving computerized tomography images collected from Erzurum City Hospital Emergency Department for the diagnosis of COVID-119 after the necessary permissions were taken. This new data set consists of 1081 lung CT images, 568 of which belong to 313 people infected with COVID-19. Later, a binary classification (Covid+/Covid-) was performed on this dataset by using MobileNetV2 and Resnet101V2 Transfer Learning (TL) architectures. In addition, the results were compared with those of other TL architectures (DenseNet121, ResNet50V2, ResNet152V2 and MobileNetV3Large). Precision, recall and f1-score parameters were used while evaluating the results. In our dataset, the highest performance values were obtained by MobileNetV2 model as follows:, %86.63 accuracy, %86.96 precision, %87.72 recall and %87.34 f1-score.The classification accuracy values achieved by other architectures are as follows: %84.79 (ResNet101V2), %84.33 (DenseNet121), %83.87 (ResNet50V2), %79.72 (ResNet152V2) and %74.65 (MobileNetV3Large). The comparison of the results showed that the highest classification accuracy value was achieved by MobileNetV2 architecture in our dataset.