eduzhai > Applied Sciences > Engineering >

Multimodality Biomedical Image Registration using Free Point Transformer Networks

  • king
  • (0) Download
  • 20210507
  • Save

... pages left unread,continue reading

Document pages: 10 pages

Abstract: We describe a point-set registration algorithm based on a novel free pointtransformer (FPT) network, designed for points extracted from multimodalbiomedical images for registration tasks, such as those frequently encounteredin ultrasound-guided interventional procedures. FPT is constructed with aglobal feature extractor which accepts unordered source and target point-setsof variable size. The extracted features are conditioned by a shared multilayerperceptron point transformer module to predict a displacement vector for eachsource point, transforming it into the target space. The point transformermodule assumes no vicinity or smoothness in predicting spatial transformationand, together with the global feature extractor, is trained in a data-drivenfashion with an unsupervised loss function. In a multimodal registration taskusing prostate MR and sparsely acquired ultrasound images, FPT yieldscomparable or improved results over other rigid and non-rigid registrationmethods. This demonstrates the versatility of FPT to learn registrationdirectly from real, clinical training data and to generalize to a challengingtask, such as the interventional application presented.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...