eduzhai > Applied Sciences > Engineering >

COVID TV-UNet Segmenting COVID-19 Chest CT Images Using Connectivity Imposed U-Net

  • king
  • (0) Download
  • 20210505
  • Save

... pages left unread,continue reading

Document pages: 9 pages

Abstract: The novel corona-virus disease (COVID-19) pandemic has caused a majoroutbreak in more than 200 countries around the world, leading to a severeimpact on the health and life of many people globally. As of mid-July 2020,more than 12 million people were infected, and more than 570,000 death werereported. Computed Tomography (CT) images can be used as an alternative to thetime-consuming RT-PCR test, to detect COVID-19. In this work we propose asegmentation framework to detect chest regions in CT images, which are infectedby COVID-19. We use an architecture similar to U-Net model, and train it todetect ground glass regions, on pixel level. As the infected regions tend toform a connected component (rather than randomly distributed pixels), we add asuitable regularization term to the loss function, to promote connectivity ofthe segmentation map for COVID-19 pixels. 2D-anisotropic total-variation isused for this purpose, and therefore the proposed model is called "TV-UNet ".Through experimental results on a relatively large-scale CT segmentationdataset of around 900 images, we show that adding this new regularization termleads to 2 gain on overall segmentation performance compared to the U-Netmodel. Our experimental analysis, ranging from visual evaluation of thepredicted segmentation results to quantitative assessment of segmentationperformance (precision, recall, Dice score, and mIoU) demonstrated greatability to identify COVID-19 associated regions of the lungs, achieving a mIoUrate of over 99 , and a Dice score of around 86 .

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×