eduzhai > Applied Sciences > Engineering >

Uniformizing Techniques to Process CT scans with 3D CNNs for Tuberculosis Prediction

  • king
  • (0) Download
  • 20210505
  • Save

... pages left unread,continue reading

Document pages: 12 pages

Abstract: A common approach to medical image analysis on volumetric data uses deep 2Dconvolutional neural networks (CNNs). This is largely attributed to thechallenges imposed by the nature of the 3D data: variable volume size, GPUexhaustion during optimization. However, dealing with the individual slicesindependently in 2D CNNs deliberately discards the depth information whichresults in poor performance for the intended task. Therefore, it is importantto develop methods that not only overcome the heavy memory and computationrequirements but also leverage the 3D information. To this end, we evaluate aset of volume uniformizing methods to address the aforementioned issues. Thefirst method involves sampling information evenly from a subset of the volume.Another method exploits the full geometry of the 3D volume by interpolatingover the z-axis. We demonstrate performance improvements using controlledablation studies as well as put this approach to the test on the ImageCLEFTuberculosis Severity Assessment 2019 benchmark. We report 73 area under curve(AUC) and binary classification accuracy (ACC) of 67.5 on the test set beatingall methods which leveraged only image information (without using clinicalmeta-data) achieving 5-th position overall. All codes and models are madeavailable at this https URL.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×