eduzhai > Applied Sciences > Engineering >

A coarse-to-fine framework for unsupervised multi-contrast MR image deformable registration with dual consistency constraint

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 11 pages

Abstract: Multi-contrast magnetic resonance (MR) image registration is useful in theclinic to achieve fast and accurate imaging-based disease diagnosis andtreatment planning. Nevertheless, the efficiency and performance of theexisting registration algorithms can still be improved. In this paper, wepropose a novel unsupervised learning-based framework to achieve accurate andefficient multi-contrast MR image registrations. Specifically, an end-to-endcoarse-to-fine network architecture consisting of affine and deformabletransformations is designed to improve the robustness and achieve end-to-endregistration. Furthermore, a dual consistency constraint and a new priorknowledge-based loss function are developed to enhance the registrationperformances. The proposed method has been evaluated on a clinical datasetcontaining 555 cases, and encouraging performances have been achieved. Comparedto the commonly utilized registration methods, including VoxelMorph, SyN, andLT-Net, the proposed method achieves better registration performance with aDice score of 0.8397 in identifying stroke lesions. With regards to theregistration speed, our method is about 10 times faster than the mostcompetitive method of SyN (Affine) when testing on a CPU. Moreover, we provethat our method can still perform well on more challenging tasks with lackingscanning information data, showing high robustness for the clinicalapplication.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...