eduzhai > Applied Sciences > Engineering >

DRR4Covid Learning Automated COVID-19 Infection Segmentation from Digitally Reconstructed Radiographs

  • king
  • (0) Download
  • 20210507
  • Save

... pages left unread,continue reading

Document pages: 21 pages

Abstract: Automated infection measurement and COVID-19 diagnosis based on Chest X-ray(CXR) imaging is important for faster examination. We propose a novel approach,called DRR4Covid, to learn automated COVID-19 diagnosis and infectionsegmentation on CXRs from digitally reconstructed radiographs (DRRs). DRR4Covidcomprises of an infection-aware DRR generator, a classification and orsegmentation network, and a domain adaptation module. The infection-aware DRRgenerator is able to produce DRRs with adjustable strength of radiologicalsigns of COVID-19 infection, and generate pixel-level infection annotationsthat match the DRRs precisely. The domain adaptation module is introduced toreduce the domain discrepancy between DRRs and CXRs by training networks onunlabeled real CXRs and labeled DRRs together.We provide a simple but effectiveimplementation of DRR4Covid by using a domain adaptation module based onMaximum Mean Discrepancy (MMD), and a FCN-based network with a classificationheader and a segmentation header. Extensive experiment results have confirmedthe efficacy of our method; specifically, quantifying the performance byaccuracy, AUC and F1-score, our network without using any annotations from CXRshas achieved a classification score of (0.954, 0.989, 0.953) and a segmentationscore of (0.957, 0.981, 0.956) on a test set with 794 normal cases and 794positive cases. Besides, we estimate the sensitive of X-ray images in detectingCOVID-19 infection by adjusting the strength of radiological signs of COVID-19infection in synthetic DRRs. The estimated detection limit of the proportion ofinfected voxels in the lungs is 19.43 , and the estimated lower bound of thecontribution rate of infected voxels is 20.0 for significant radiologicalsigns of COVID-19 infection. Our codes will be made publicly available atthis https URL.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...