Quantifying Productivity gains from using Artificial Intelligence (AI) in detecting COVID-19 patterns from Chest X-Rays

A. Background Coronavirus disease (COVID-19) presents with non-specific respiratory symptoms that vary in severity and range from mild, severe to life threatening conditions requiring advanced mechanical respiratory support. Currently, identification of viral RNA in reverse transcriptase polymerase chain reaction (RTPCR) is regarded as the gold standard diagnostic test for COVID-19. However, RT-PCR has been shown to have limitations such as high number of false negatives and delayed results more especially in resource limited settings. Chest imaging has been used to complement clinical evaluation and laboratory workup in diagnosis and management of patients highly suspected or confirmed to have COVID-19 and most centers have reported literature on Chest CT manifestations in COVID-19 compared to other imaging modalities. In fact, there is developing literature identifying higher sensitivity of Chest CT for diagnosis of COVID-19 as compared with initial RT-PCR from swab samples. However, due to some limitations of CT in terms of infection control, availability in resource-limited areas, portability, some centers have used Chest radiography (CXR) and Lung ultrasound (LUS) to identify lung abnormalities pertinent to COVID-19. B. Goals and Objectives The overall objective of this study is to quantify potential productivity gains from use of Artificial Intelligence (AI) in detecting COVID-19 patterns from Chest X-Rays. Specific objectives are: i) to build highly accurate AI models capable of detecting COVID-19 patterns from Chest X-Ray, as well as ii) (ii) to create a setup that would allow us to measure improvements in speed of diagnosis and accuracy by radiologists assisted by AI in diagnosing COVID-19 from Chest X-Ray. C. Methods Researchers shall perform a retrospective and analytical study in which Chest X-Ray images of confirmed COVID-19 patients are studied by radiologists to detect the patterns in the image that are typical to COVID-19 infection. The first component of the study will comprise image annotation steps and will be done by expert radiologists based in Kigali, Rwanda using Insightiv's Teleradiology platform (i.e. online). They will use Chest-X-ray images collected by combining open-source and proprietary data from different partner institutions around the world. Phase 2 and 3 of the study will be done by expert software engineers and machine learning engineers based in Kigali, Rwanda, using cloud-based Graphical Processing Units (GPUs) to train different algorithms on detecting COVID-19 patterns from images annotated by radiologists, as well as on evaluating the algorithms. D. Expected outcomes By leveraging AI and Chest X-Ray imaging modality, the study will provide a cost-effective technology for risk assessment and early isolation of COVID-19 suspects, a win-win for suspected patients who are isolated and treated early without spreading the disease, and for the healthcare system which is overwhelmed with high cost and volume of patients