UNBIASED NONLINEAR IMAGE REGISTRATION
We present a novel framework for constructing large deformation unbiased image registration models that generate theoretically and intuitively correct deformation maps. Unbiased registration models do not rely on regridding and are inherently diffeomorphic and topology preserving. To demonstrate the power of the proposed framework, we generalize the well known viscous fluid registration model to compute log-unbiased deformations. We tested the proposed method using a pair of binary corpus callosum images, a pair of two-dimensional serial MRI images, and a set of three-dimensional serial MRI brain images. We compared our results to those computed using the viscous fluid registration method, and demonstrated that the proposed method is advantageous when recovering voxel-wise maps of local tissue change.
| template image | study image | deformed template |
 |
 |
 |
| THREE-DIMENSIONAL SERIAL MRI VOLUMES Click on any image to see the volume deforming. |
displacement field |
Jacobian map of the deformation |
 |
 |
|
|
Three-dimensional volume is deformed using the proposed unbiased nonlinear registration model. |
||
| fluid registration | unbiased fluid registration | ||
|
CORPUS CALLOSUM IMAGES Click on any image to see details. |
 |
 |
|
|
The generated grids are superimposed with the deformed images. Blue and red contours represent the boundaries of ventricles in the template and deformed template, respectively. The first image shows grid lines merge and self-cross for the fluid registration model, which is a consequence of negative Jacobian values at certain places, indicating topology change. The resulting grid of the proposed method is more regular. |
|||
| Click on any image to see details. |
 |
 |
 |
|
Jacobian map of the deformation is superimposed with the deformed image for the fluid registration model and the proposed model. |
|||
| fluid registration | unbiased fluid registration | ||
|
TWO-DIMENSIONAL SERIAL MRI IMAGES Click on any image to see details. |
 |
 |
|
| Click on any image to see details. |
 |
 |
|
|
The generated grids are superimposed with the deformed images (first row). Blue and red contours represent the boundaries of ventricles in the template and deformed template, respectively. The resulting grid of the proposed method is visually more regular. The deformed grids are shown separately on the second row. |
|||
| Click on any image to see details. |
 |
 |
 |
|
Jacobian map of the deformation is superimposed with the deformed image for the fluid registration model and the proposed model. |
|||
| THREE-DIMENSIONAL SERIAL MRI VOLUMES | |||
| fluid registration | |||
 |
 |
 |
 |
|
unbiased fluid registration |
|||
|
|
|
|
|
|
axial plane |
axial plane |
coronal plane |
sagittal plane |
|
|
|||
|
Jacobian maps are superimposed with the deformed volumes for the fluid registration model (row 1) and the proposed unbiased image registration model (row 2). Columns depict slices in axial (columns 1 and 2), sagittal (column 3), and coronal (column 4) planes. Right temporal atrophy (RT) and ventricular enlargement (V) are easily visualized in the Jacobian map generated using the proposed method, while the viscous fluid method generated a very noisy map. |
|||
References:
Igor Yanovsky, Paul Thompson, Stanley Osher, Alex Leow, Topology Preserving Log-Unbiased Nonlinear Image Registration: Theory and Implementation, IEEE Conference on Computer Vision and Pattern Recognition, June 2007.
Alex Leow, Igor Yanovsky, Ming-Chang Chiang, Agatha Lee, Andrea Klunder, Allen Lu, James Becker, Simon Davis, Arthur Toga, Paul Thompson, Statistical Properties of Jacobian Maps and the Realization of Unbiased Large-Deformation Nonlinear Image Registration, IEEE Transactions on Medical Imaging, vol. 26, no. 6, pp. 822-832, 2007.
