Article Preview
TopIntroduction
Coronavirus disease (COVID-19), caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), saw its first outbreak in early December 2019. Since then, the virus quickly spread throughout the globe and was declared as a global pandemic in March 2020. Up until this study, more than 184 million cases of the virus and over 3.9 million fatalities have been reported worldwide. A key step in curbing the spread of the virus and reducing the number of fatalities, is the early detection of the disease. One of the most widely used methods for detecting the virus are reverse polymerase chain reaction (RT-PCR) tests or other nucleic acid tests on samples collected from a nasopharyngeal swab. While they have played a vital role in obtaining a diagnosis for the disease, RT-PCR tests are known to have a low sensitivity rate and are prone to more false negatives due to milder cases and probable human error (Ai et al., 2020).
Further, recent studies are being conducted to understand the efficiency of Chest CTs as means for obtaining a diagnosis and upon inspecting the efficacies of Chest-CT and RT-PCR tests for the same, Xie et al. (2020) found that Chest-CT scans were able to detect positive samples for 75% of negative RT-PCR samples. Another study conducted by Fang et al. (2020), assessed the difference in the sensitivity rates between these two methods and found that the sensitivity rates of Chest-CTs and RT-PCR tests were calculated to be 98 and 71%, respectively, indicating that using radiology scans as a means of diagnosing the virus showed good promise and is also a faster means of diagnosis. However, manual evaluation of the overwhelmingly large number of X-ray images during this pandemic is time-consuming and labour-intensive, as well as subject to human interpretation and error. Therefore, there is an increasing demand to better the efficiency of this process and design computer-aided diagnostic systems to provide a reliable solution to remedy these issues. Deep learning (DL) techniques allow machines to identify, extract and learn the hidden abnormalities in the radiology images that can be missed by other Machine Learning (ML) techniques, allowing us to use algorithms that eliminate the need to manually process and analyse data (Ghosh et al, 2020). Thus, making use of these techniques with radiology images in conjunction with clinical observations together can help to provide a timely diagnosis of the disease.