Hierarchical Automatic COVID-19 Detection via CT Scan Images

Abstract

The novel coronavirus disease (COVID-19) had its outbreak in December 2019. It has since spread across the world and caused great loss of life. Nowadays, computer tomography (CT) scans are a common and effective tool to detect COVID-19. However, manually detecting a huge amount of CT scans adds great pressure and causes additional workloads for physicians and radiologists, especially for those in areas where there is a severe COVID-19 pandemic. Driven by the desire of alleviating a medical worker’s burden, here, we propose a hierarchical method in COVID-19 detection via CT scans in order to obtain a much faster detection result and one that is less labor-intensive. In this study, we present an automatic COVID-19 detection method, which consists of a hierarchical model made-up of two stages: a segmentation stage followed by a classification stage. In the segmentation stage, a U-Net is used to segment the lung portion from chest CT slices in order to eliminate the interference of irrelevant tissues such as the heart and bones. In the classification stage, ResNet-18 is applied to classify previously segmented CT slices (from the previous stage) and predict the existence of COVID-19. Experimental results show that our proposed hierarchical detection method obtains satisfying performances in separating COVID-19 CT scans from common pneumonia CT scans at the scan level, indicating that the method has great potential in assisting physicians and radiologists in rapid COVID-19 detection and significantly reducing their workload.

Ähnliche Arbeiten