3BResNet: A Novel Residual Block-Based ResNet Model Approach for COVID19 Detection

Abstract

In recent years, upper respiratory tract infections that have affected the whole world have caused the death of millions of people. It is predicted that similar infections may occur in the coming years. Therefore, it is necessary to develop methods that can be used widely, especially during epidemic periods. The study developed a decision support system for use in upper respiratory tract infections. At this stage, first, the ResNet models in the literature were examined and an application was developed on the SARS-CoV-2 Ct dataset. Next stage, the block structure in the ResNet models in the literature was changed, the number of layers was reduced, and a new model was proposed that provides higher success with fewer parameters. With the proposed model, the values 0.97, 0.97, 0.94, and 0.98 were achieved for accuracy, F1 score, precision and sensitivity on the SARS-CoV-2 Ct dataset, respectively. When the obtained values are compared to state of the art methods in the literature, it has been determined that they are at a competitive level with much fewer parameters. Hardware-related problems encountered in the training of ResNet models at low hardware levels were solved with the proposed model, resulting in a higher success rate. Furthermore, the proposed model can be widely used in different decision support systems that are urgently needed in adverse conditions such as pandemics due to its lightweight structure and high-performance results. As a result of the study, a new model that can provide higher performance with much lower layer structure than existing ResNet models has been introduced into the literature with the proposed model.