Enhancement of Covid-19 Segmentation Using Machine Learning Analyses of Lung Imaging

ABSTRACT

The Covid-19 pandemic has caused more then 193 million cases and 4.1 million deaths worldwide as of July 2021. The Fleischner Society reported that Computerized Tomography (CT) is a useful tool for the early identification of Covid-19. Covid-19 disease induces lung changes which can be observed in lung CT predominantly as ground-glass opacification (GGO) and occasional consolidation in the peripheries. Moreover, it was reported that the percentage of lung showing disease correlates with the severity of the disease. Therefore, segmentation of the disease areas in CT images is a logical first step to quantify disease severity. In this paper, we propose g CoviSegNet Enhanced' based on a U-Net with an 813-layer EfficientNetB7 encoder having an attention mechanism to segment the Covid-19 disease area observed in CT images of Covid-19 patients. CoviSegNet Enhanced is an improvement of our previous work g CoviSegNet'. The experiments performed on three public CT datasets and a detailed comparison with recently published work confirms that the proposed CoviNet Enhanced using deep learning approaches is highly effective for Covid-19 segmentation. © 2021 ACM.