Deformable Image Registration with Inclusion of Autodetected Homologous Tissue Features

1
Deformable Image Registration with Inclusion of
Auto-detected Homologous Tissue Features

• Yao Xie, Lei Xing
• Department of Radiation Oncology
• Stanford University School of Medicine

2
Purpose

• To implement an automatic image registration,
  which can distinguish the rigid tissue and the
  soft tissue.

3
Two-Step Image Registration based on Control
Volumes

• First step auto-map control volumesRigid
  registration
• Second step thin plate spline (TPS) Deformable
  registration

4
Improvements

• From auto-map control volume to auto-search
  control volume
• From auto-search control volume to auto-search
  feature region
• From single-directional mapping to bi-directional
  mapping

5
Feature region
6
(No Transcript)
7
Descriptor

• Descriptor has 3 parameters (x,y,?).
• Orientation histogram of gradient magnitudes

8
2D Rigid Registration


Fixed image
Moving image
9
2D Rigid Registration

10
2D Deformable Registration



Moving image
Fixed image
11
2D Deformable Registration


12
4D CT Registration
13
3D Deformable Registration
14
Axial view
15
Frontal view
16
Sagittal view
17
Conclusions

• Registration is extended from control volumes to
  arbitrary shapes, which extends the voxel-based
  or volume-based method to tissue-based method,
  and separate the rigid tissues from deformable
  tissues in an image.
• Develop the auto-search method for feature region
  to avoid the manually search for control point.