Detection of COVID-19 Disease with Machine Learning Algorithms from CT Images

Abstract

With the identification of the SARS-COV-2 virus in November 2019, the world has become very different. The COVID-19 disease caused by the virus has reached epidemic proportions and continues. This virus, which is one of the most contagious and deadly pathogens in human history with the number of cases approaching 600 million and the number of deaths exceeding 6 million, has shown and continues to show itself in every area that people come into contact with, from business life to economy, transportation to education, social life to psychology. Although the developed vaccines provide a partial decrease in the number of deaths, the mutations that the virus constantly undergoes and the increase in the transmission rate accordingly reduce the effectiveness of the vaccines, and the number of deaths tends to increase as the number of infected people. It is undoubtedly important that the detection of this epidemic disease, which is the biggest crisis that humanity has experienced in the last century after World War II, is carried out accurately and quickly. In this study, a machine learning-based artificial intelligence method has been proposed for the detection of COVID-19 from computed tomography images. The features of images with two classes are extracted using the Local Binary Pattern. The images reserved for training in the dataset were used for training machine learning models. Trained models were tested with previously unused test images. While the Fine K-Nearest Neighbors model reached the highest accuracy with a value of 0.984 for the training images, the highest accuracy value was obtained by the Cubic Support Vector Machine with 0.93 for the test images. These results are higher than the deep learning-based study using the same data set.