Interventional 2D/3D Registration with Contextual Pose Update
Embargo until
Date
2019-05-06
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Johns Hopkins University
Abstract
Traditional intensity-based 2D/3D registration requires near-perfect initialization
in order for image similarity metrics to yield meaningful gradient updates
of X-ray pose. They depend on image appearance rather than content, and
therefore, fail in revealing large pose offsets that substantially alter the appearance
of the same structure. We complement traditional similarity metrics with
a convolutional neural network-based (CNN-based) similarity function that
captures large-range pose relations by extracting both local and contextual information,
and proposes meaningful X-ray pose updates without the need for
accurate initialization. Our CNN accepts a target X-ray image and a digitally
reconstructed radiograph at the current pose estimate as input and iteratively
outputs pose updates on the Riemannian Manifold. It integrates seamlessly
with conventional image-based registration frameworks. Long-range relations
are captured primarily by our CNN-based method while short-range
offsets can be recovered accurately with an image similarity-based method.
On both synthetic and real X-ray images of the pelvis, we demonstrate that the
proposed method can successfully recover large rotational and translational
offsets, irrespective of initialization.
Description
Keywords
Image-guided surgery, machine learning, X-ray, CT, Riemannian Manifold