ITK Release 4/A2D2 Projects/Physics-Based Non-rigid Registration

From KitwarePublic
< ITK Release 4‎ | A2D2 Projects
Revision as of 19:21, 2 September 2011 by Kot (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigationJump to search

Goals

To develop an ITK implementation of physics-based Non-Rigid Registration (NRR) for Image- Guided Surgery (IGS) that will satisfy the following requirements: account for tissue properties in the registration, improve accuracy compared to rigid registration, and reduce execution time to less than one minute.

Motivation

The benefits to the ITK community are at least two-fold. First, as a stand-alone software we will provide a computationally efficient registration method that accounts for tissue properties and that approximates sparse deformation. Second, through ITK integration into other IGS toolkits, NRR will be part of open-source systems like 3D-Slicer and IGSTK, and commercial systems like BrainLab.

Methodology

Our methodology is based on the separation of the NRR method into two parts: a regular part, block matching that utilizes the GPU, and an irregular part, a Finite Element solver that is mapped to multi-core processors

Deliverable

Feature Point Selection filter as itk::MaskFeaturePointSelection3D derived from itk::ImageToMeshFilter

Input:

  • floating image as `itk::Image`
  • mask image as `itk::Image`

Output:

  • list of selected points as itk::PointSet

For each point we store:

  • image coordinates
  • real coordinates
  • tensor


Block Matching filter as itk::BlockMatching3D

Input:

  • floating image as itk::Image
  • fixed image as itk::Image
  • list of selected points (output of itk::MaskFeaturePointSelection3D) as itk::PointSet

Output:

  • displacement field as itk::PointSet

For each point we store:

  • image coordinates
  • real coordinates
  • tensor
  • confidence
  • displacement
  • correlation


FEMScatteredDataPointSetToImageFilter

Description:

FEM-based approximation to interpolation filter. The scattered point set is approximated by a piece-wise linear function regularized by a linear elastic biomechanical model. The approximation error is reduced iteratively to reach interpolation accuracy. This filter takes a point set, mesh and material as inputs and outputs a deformation field. The following figures show the flow chart and the inheritance diagram, respectively.

FEMScatteredDataPointSetToImageFilter.jpg


RobustSolver

Description:

A robust solver against outliers. This solver is developed based on the interface of LinearSystemWrapper, which can be implemented by vnl or PETSc.

RobustSolver.jpg

Team

  • Nikos Chrisochoides ( ODU )
  • Andriy Kot ( ODU )
  • Yixun Liu ( ODU )
  • Michel Audette ( ODU )
  • Andinet Enquobahrie ( Kitware )