Automated lung segmentation from CT images of normal and COVID-19\n pneumonia patients

Abstract

Automated semantic image segmentation is an essential step in quantitative\nimage analysis and disease diagnosis. This study investigates the performance\nof a deep learning-based model for lung segmentation from CT images for normal\nand COVID-19 patients. Chest CT images and corresponding lung masks of 1200\nconfirmed COVID-19 cases were used for training a residual neural network. The\nreference lung masks were generated through semi-automated/manual segmentation\nof the CT images. The performance of the model was evaluated on two distinct\nexternal test datasets including 120 normal and COVID-19 subjects, and the\nresults of these groups were compared to each other. Different evaluation\nmetrics such as dice coefficient (DSC), mean absolute error (MAE), relative\nmean HU difference, and relative volume difference were calculated to assess\nthe accuracy of the predicted lung masks. The proposed deep learning method\nachieved DSC of 0.980 and 0.971 for normal and COVID-19 subjects, respectively,\ndemonstrating significant overlap between predicted and reference lung masks.\nMoreover, MAEs of 0.037 HU and 0.061 HU, relative mean HU difference of -2.679%\nand -4.403%, and relative volume difference of 2.405% and 5.928% were obtained\nfor normal and COVID-19 subjects, respectively. The comparable performance in\nlung segmentation of the normal and COVID-19 patients indicates the accuracy of\nthe model for the identification of the lung tissue in the presence of the\nCOVID-19 induced infections (though slightly better performance was observed\nfor normal patients). The promising results achieved by the proposed deep\nlearning-based model demonstrated its reliability in COVID-19 lung\nsegmentation. This prerequisite step would lead to a more efficient and robust\npneumonia lesion analysis.\n