2020/07/23 by Narges Saeedizadeh, Shervin Minaee, Saeedizadeh, Narges +7
Medicine · #COVID-19 diagnosis using AI #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #FOS: Electrical engineering #Image and Video Processing (eess.IV) #Machine Learning (cs.LG) #Medical Imaging Techniques and Applications #Radiomics and Machine Learning in Medical Imaging #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2007.12303
openalex publication_date 2020/07/23 · openalex created_date 2022/07/26 · openalex updated_date 2026/07/28
The novel corona-virus disease (COVID-19) pandemic has caused a major\noutbreak in more than 200 countries around the world, leading to a severe\nimpact on the health and life of many people globally. As of mid-July 2020,\nmore than 12 million people were infected, and more than 570,000 death were\nreported. Computed Tomography (CT) images can be used as an alternative to the\ntime-consuming RT-PCR test, to detect COVID-19. In this work we propose a\nsegmentation framework to detect chest regions in CT images, which are infected\nby COVID-19. We use an architecture similar to U-Net model, and train it to\ndetect ground glass regions, on pixel level. As the infected regions tend to\nform a connected component (rather than randomly distributed pixels), we add a\nsuitable regularization term to the loss function, to promote connectivity of\nthe segmentation map for COVID-19 pixels. 2D-anisotropic total-variation is\nused for this purpose, and therefore the proposed model is called "TV-UNet".\nThrough experimental results on a relatively large-scale CT segmentation\ndataset of around 900 images, we show that adding this new regularization term\nleads to 2 % gain on overall segmentation performance compared to the U-Net\nmodel. Our experimental analysis, ranging from visual evaluation of the\npredicted segmentation results to quantitative assessment of segmentation\nperformance (precision, recall, Dice score, and mIoU) demonstrated great\nability to identify COVID-19 associated regions of the lungs, achieving a mIoU\nrate of over 99 %, and a Dice score of around 86 %.\n