Investigation of the performance of data mining and deep learning methods for COVID-19 positive detection in lung graph with medical image processing

The World Health Organization (WHO) declared the novel coronavirus (COVID-19), a contagious epidemic caused by SARS-CoV-2, as a pandemic in March 2020. In all countries affected by this virus, the number of infected and deceased patients is increasing day by day. It is important to have smart predictive and diagnostic tools, as early detection can reduce the spread of the virus. Creating effective predictive models can help in implementing better health strategies to prevent the spread of the virus. Today, with the help of medical images, artificial intelligence-based solutions are used in the detection of many health problems (brain tumor detection, breast cancer detection, pneumonia detection, etc). In this pandemic process we are in, many artificial intelligence-based studies have been carried out to shorten the diagnosis time of COVID-19 and to increase the correct diagnosis rate. In this study, chest radiographs were digitized using medical image processing techniques and five different image processing methods were applied to examine the effects of these methods on the classification performance of chest radiographs. In addition, feature extraction was carried out with convolutional neural networks, which is a deep learning technique, and the effect on the classification of images with SVM, RF, ANN classical data mining methods was compared. The open access COVID-19 Radiography database was used in the study. The study was planned with 3616 COVID-19, 6012 Non-COVID and 10192 Normal chest radiographs in this data set. It has been found that image processing techniques have an effect on the classification performance of chest radiographs. After the features were extracted with the convolutional neural network method, the classification performances obtained by classical data mining methods showed better results than the classification performances obtained only by classical data mining.