Method for Deformable 3D-2D Registration Using Multiple Locally Rigid Registrations

Case ID:
C13844
Disclosure Date:
10/12/2015
Description:
Unmet Need
3D preoperative imaging (e.g., CT, MRI, PET) provides the basis for many forms of surgical planning and intraoperative guidance. During the surgical planning stage, clinicians may define geometric annotations (points, contours, shapes, etc.) in the preoperative 3D images, such as outlining of relevant anatomical structures and identification of desired placement/trajectory of surgical hardware. These annotations can be incorporated into the surgical process using various algorithms for rigid/deformable registration (aligning physical coordinate systems to create a “mapping” from one coordinate system to another).  In the context of surgical guidance, it is often useful to create such a mapping between the 3D preoperative images and the 2D intraoperative images (e.g., x-ray radiographic / fluoroscopy systems). This mapping then enables the annotations defined in the 3D image to be overlaid onto the 2D image, providing decision support for the clinician as well as a means for verification of the surgical product.
Existing methods for image-intensity-based 3D-2D annotation mapping often use a rigid transformation and can be limited by deformation in anatomy. These deformations occur commonly due to differences in patient positioning between the preoperative acquisition and intraoperative acquisition during surgery. Other sources of deformation include patient motion, breathing, or the procedure itself – e.g., correction of spinal curvature. For example, 3D images are often acquired when the patient lying in a supine position (on the CT scanner table), whereas surgery often requires the patient to lie in a prone position (on the OR table). This results in a deformation of anatomy, including the structures of interest in surgery (e.g., the spinal vertebrae).  Deformable registration could improve accuracy in these scenarios, but such methods are susceptible to local optima and often fail due to a large number of parameters being optimized. Piece-wise rigid methods exist in feature-based registration to account for these deformations; however, segmentations or shape models need to be extracted from the 3D image to perform the registration, creating additional work in planning, introducing sources of error in segmentation, and often disregarding potentially relevant image intensity information.
Technology Overview
The invention comprises a multi-stage masking approach to intensity-based 3D-2D registration. At the first stage, the entire 3D image is (optionally) masked in a manner that includes all annotations and structures of interest – the same as the previously reported LevelCheck algorithm – and provides initialization to subsequent stages. At each subsequent stage, the images are divided into fragments, based on subsets of the annotations in the 3D image, to perform multiple local registrations, thus minimizing the impact that global deformations may have in an entirely rigid registration. This solution is analogous to block-matching, proposed for video motion correction applications and other 2D-2D / 3D-3D registration methods. However in this application, rather than arbitrarily dividing the image to perform separate registrations and then combining the outputs to create a deformation field or a single rigid registration, the proposed method defines the “blocks” according to the annotations that have been defined in the 3D image, and computes multiple registrations that are accurate within the local region of associated annotations. This can then act as a refinement to the existing rigid 3D-2D registration to provide improved accuracy without requiring additional user input.
Stage of Development: 
Calibration and validation results demonstrate reliable measurement of the contact force as well as location of the sclerotomy. Preliminary experiments have been conducted to functionally evaluate robotic intraocular illumination.
 
Publications: 
Phys Med Biol. 2016 Apr 21; 61(8): 3009–3025.
 
Patent Information:
Title App Type Country Serial No. Patent No. File Date Issued Date Expire Date Patent Status
Method for Deformable 3D-2D Registration Using Multiple Locally Rigid Registrations ORD: Ordinary Utility United States 15/381,494 12/16/2016     Pending
Inventors:
Category(s):
For Information, Contact:
Jon Gottlieb
jgottl10@jhu.edu
410-614-0300
Save This Technology:
2017 © Johns Hopkins Technology Ventures. All Rights Reserved. Powered by Inteum