Diagnosing COVID-19 disease using artificial intelligence methods

A new process has started with Covid 19 for artificial intelligence and especially machine learning processes. Since the coronavirus epidemic is so widespread and deadly, it is of great importance to detect the coronavirus using computer-aided systems. In this thesis, a CNN-based hybrid model has been developed. Resnet101 and Densenet201 architectures were used as the base in the model developed to detect the coronavirus disease. Feature maps were obtained from the fc1000 layers of the Resnet101 and Densenet201 architectures. The size of each feature map obtained from the Resnet101 and Densenet201 architectures is 3075 x 1000. These feature maps obtained using the Resnet101 and Densenet201 architectures were then combined. The size of the combined feature map was 3075 x 2000. In this way, different features of the same image are brought together. This will increase the performance of the developed model. In the last step of the proposed hybrid model, the combined feature maps are classified in the SVM classifier. In this thesis study, different pre-trained CNN architectures accepted in the literature were used to test the performance of the proposed model. As a result, the proposed hybrid model has been more successful than the CNN architectures accepted in the literature. The proposed hybrid model achieved an accuracy of 98.2%. It is obvious that the results obtained in the proposed model can be used for pre-diagnosis in non-experts, and that the proposed model will alleviate the workload of experts and reduce costs in specialists.