Smart Diagnostics of COVID-19 With Data-Driven Approaches

Introduction

Since the first SARS-CoV-2 confirmed case was found in Wuhan, China in late December 2019, such a novel coronavirus and its variants have been sweeping around the world for over 18 months as of July 2021. COVID-19 diagnosis plays a critical role in the containment of the disease in terms of the policy-making process through infected case identification, quarantine, and contact tracking. As a respiratory and infectious disease, the regular medical diagnosis of COVID-19 primarily depends on tracking epidemiological history and characteristics (Chang et al., 2020), analyzing clinical symptoms (Li et al., 2020), and confirmed by a variety of medical identification processes, such as nucleic acid amplification tests (Basu et al., 2020), various medical imaging tests including computed tomography (CT) scans (Awulachew et al., 2020), chest x-rays (Durrani et al., 2020), and magnetic resonance imaging (MRI) tests (Langenbach et al., 2020), and a variety of biological tests, i.e. serological assays (Krammer & Simon, 2020), protein tests (Poggiali et al., 2020), enzyme-linked immunoassay (Xiang et al., 2020), and lateral flow antigen detections (Wu et al., 2020a). Early epidemiological studies uncover the clinical characteristics of the novel coronavirus and its infection trails, which have concluded that the COVID-19 diagnostic result cannot be determined solely on the identification of common symptoms, such as fever, cough, difficulty of breath, and fatigue (Guan et al., 2020; Huang et al., 2020). Many of such symptoms can be associated with other respiratory diseases, thereby are not specific and may not be adopted for the precise diagnosis. Besides, the total number of reported infections is underestimated as many mild and asymptomatic cases are not included, whereas such cases are still difficult to detect with a symptoms-based diagnosis (Kobayashi et al., 2020). Therefore, the accurate diagnosis of SARS-CoV-2 is still challenging as the clinical manifestation of such a novel virus is significantly variable from case to case, particularly, with asymptomatic to acute respiratory distress syndrome and multi-organ failures (Girija et al., 2020; Zaim et al., 2020).

With the development of genetics, molecular techniques, and medical image processing, exploring the biological properties of a disease and diagnostic approaches becomes more accurate and efficient. Such ideologies and technologies have been applied to understand the novel coronavirus. Early genetic studies indicate that SARS-CoV-2 is a single-stranded positive RNA genome that encodes an RNA-dependent RNA polymerase and four structural proteins, including the spike surface glycoprotein, a small envelope protein, a matrix protein, and the nucleocapsid protein (Wu et al., 2020b; Walls et al., 2020). A negative result based on real-time polymerase chain reaction (PCR) from the experimental design using nucleic acid test suggests that the origin of the cause of pneumonia is unclear compared with the known pathogen panels (Zhou et al., 2020). Further biometric studies discover that a pathogen with a similar genetic sequence to the beta coronavirus B lineage matches the genome of the bat coronavirus RaTG13, the severe acute respiratory syndrome virus (SARS-CoV), and the Middle East respiratory syndrome virus (MERS-CoV), whereas the similarity of each genome varies from 50% to 96% (Lu et al., 2020). The initial diagnosis of Covid-19 has been led by scanning patients using CT scan. Clinical findings from CT images reveal the differences between healthy lungs and lungs with Covid-19, which are denser, more profuse, and confluent (Ai et al., 2020; Chung et al., 2020). Therefore, nucleic acid tests and initial diagnostics are still essential in the context of Covid-19, providing the knowledge graph for experts to understand the virus efficiently.