eduzhai > Applied Sciences > Engineering >

Feedback Graph Attention Convolutional Network for Medical Image Enhancement

  • Save

... pages left unread,continue reading

Document pages: 0 pages

Abstract: Artifacts, blur and noise are the common distortions degrading MRI imagesduring the acquisition process, and deep neural networks have been demonstratedto help in improving image quality. To well exploit global structuralinformation and texture details, we propose a novel biomedical imageenhancement network, named Feedback Graph Attention Convolutional Network(FB-GACN). As a key innovation, we consider the global structure of an image bybuilding a graph network from image sub-regions that we consider to be nodefeatures, linking them non-locally according to their similarity. The proposedmodel consists of three main parts: 1) The parallel graph similarity branch andcontent branch, where the graph similarity branch aims at exploiting thesimilarity and symmetry across different image sub-regions in low-resolutionfeature space and provides additional priors for the content branch to enhancetexture details. 2) A feedback mechanism with a recurrent structure to refinelow-level representations with high-level information and generate powerfulhigh-level texture details by handling the feedback connections. 3) Areconstruction to remove the artifacts and recover super-resolution images byusing the estimated sub-region correlation priors obtained from the graphsimilarity branch. We evaluate our method on two image enhancement tasks: i)cross-protocol super resolution of diffusion MRI; ii) artifact removal of FLAIRMR images. Experimental results demonstrate that the proposed algorithmoutperforms the state-of-the-art methods.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×