Deep learning from computerized tomography imagesand diagnosis of COVID-19 with machine learning

COVID-19 disease has become an epidemic that spread rapidly in a short time and caused many people to get sick and die. Rapid detection of the disease is of great importance in controlling the Covid-19 disease and reducing the transmission. Lung imaging procedures constitute an important part of the diagnosis of COVID-19 disease. In this study, a data set containing a total of 7200 images was created from the lung images of the Computed Tomography device, which has not been used in any study before. Computed Tomography images were selected from patient data reported with the help of radiology doctors and radiology technicians from all adult patients between March 2020 and November 2020. The binary classification method was used in the study. For the diagnosis of the disease, validation and testing processes were carried out in Convolutional Neural Networks from Deep Learning methods, Artificial Neural Network (ANN) from Machine Learning algorithms, Support Vector Machine (SVM), K-Nearest Neighbor (KNN) algorithms. As a result of the studies, 99.77% accuracy and 97.78% test accuracy were obtained for the CNN architecture Xception model during the training phase. For the ANN model, 98.98% accuracy and 97.68% test accuracy were obtained during the training phase. The test process performed with the SVM model achieved 96.91% accuracy. In the KNN model, the best test accuracy of 97.40% was obtained for the given K values.