eduzhai > Applied Sciences > Engineering >

Stabilizing Deep Tomographic Reconstruction Networks

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 68 pages

Abstract: Tomographic image reconstruction with deep learning (DL) is an emerging fieldof applied artificial intelligence, but a recent landmark study reveals thatseveral deep reconstruction networks are unstable for computed tomography (CT)and magnetic resonance imaging (MRI). Since deep reconstruction is now amainstream approach to achieve better tomographic image quality, stabilizingdeep networks is an urgent challenge. Here we propose an Analytic CompressiveIterative Deep (ACID) framework to address this challenge. Instead of onlyusing DL or compressed sensing, ACID consists of four modules: deep learning,compressed sensing-inspired sparsity promotion, analytic mapping, and iterativerefinement. This paper shows the convergence and stability of ACID under abounded error norm condition (a special case of the Lipschitz continuity),improves deep reconstruction quality by stabilizing an unstable deepreconstruction network in the ACID framework, and demonstrates the power ofACID in both stabilizing an unstable network and being resilient againstadversarial attacks to the whole ACID workflow. In our experiments, ACIDeliminated all three kinds of instabilities and significantly improved imagequality in the context of the aforementioned study on the instabilities,demonstrating that data-driven reconstruction can be stabilized to outperformreconstruction using sparsity-regularized reconstruction alone. The mechanismof ACID is to synergize a deep reconstruction network trained on big data,compressed sensing-based improvement with kernel awareness, and iterativerefinement to eliminate any data residual inconsistent with real data. Weanticipate that this integrative closed-loop data-driven approach helps advancedeep tomographic image reconstruction methods into clinical applications.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×