Development of deep learning based models for COVID-19 detection from X-ray images

Thousands of people died as a result of the Covid-19 epidemic, which was caused by severe acute respiratory syndrome Coronavirus 2 (SARS-CoV-2). The most critical element in combating this deadly epidemic is early diagnosis. Through early diagnosis and quarantine, it is possible to reduce the rate of transmission and death. Today, PCR (Polymerase Chain Reaction) tests and lung X-ray images are used in the diagnosis of Covid-19. X-ray images are of vital importance in early diagnosis, as the conclusion of PCR tests is a time-consuming process. The overall goal of this study is to develop a diagnostic methodology based on X-ray images in order to assist experts in diagnosing Covid-19. In the first stage of the methodology, features are extracted from the dataset consisting of X-ray images by using a pre-trained SqueezeNet architecture. In the second stage, nearest component analysis (NCA) is applied to the extracted feature set, and noisy ones are eliminated from this feature set. In the final stage, Covid-19 detection is performed on the noise-free feature set with support vector machines. The performance analysis of the proposed methodology is made by comparing it with a variety of deep pre-trained architectures and classifiers on 5 different data sets. According to the results, the proposed methodology achieves better results than others.