Covid-19 Diagnosis Based on CT Images Using Pre-Trained Models

摘要

the NOVEL (COVID-19) coronavirus has recently grown into a pandemic in the world due to the severe acute respiratory syndrome (SARSCoV-2). According to studies in this area, about 34,440,235 people are infected with COVID-19, 1,023,430 is the number of deaths, and around 25,633,956 patients are being subjected to treatment worldwide. In this paper researchers used five pre-trained models. They are: ResNet-50, ResNet-101, AlexNet, VGG11, and SqueezeNetV-1.0. DTL (deep transfer learning) is used to diagnose the NOVEL (COVID-19) by training the COVID-19 coronavirus dataset with 32-batch size and 25 epochs. In training, ResNet-50 gives the best value in loss rate (0.22) with an accuracy of 93.2%, whereas, VGG11 showed the worst value (0.38). Also, in validation, the results showed that ResNet-50 (0.28) is the best, and VGG11 achieved (0.39) as the worst value.