Builtin Function Types

Enhancement NO Segmentation NO Quantification NO Registration YES Visualization NO Compression NO

The Automated Image Registration (AIR) package is designed for two-dimensional or three-dimensional image registration. It is distributed to the imaging community as C source code available for download at no charge (see details at http:// bishopw.loni.ucla.edu/AIR3/). As distributed, the package supports the ANALYZE file format and includes utilities for converting raw images to this format. The availability of source code makes it possible to implement other file formats as well.

Automated registration is achieved using a calculus-based minimization algorithm. The implemented cost functions for registration are least squares (with or without intensity scaling) and ratio image uniformity. This latter cost function allows cross-modality registration in many instances, especially when combined with image segmentation (segmentation tools are not included as part of the package). Minimization can be constrained using a variety of linear models including rigid-body constraints, rigid-body with global rescaling constraints, affine constraints, and perspective constraints. In addition, nonlinear warping, constrained by low-order polynomials, can be performed using a least squares cost function with intensity scaling.

The package allows the derived linear transformations to be combined with one another or with polynomial warps so that a series of transformation steps can be combined into a single mathematical description. This minimizes data loss when portions of the image would lie outside the field of view in one of the intermediate reference frames. Also, combining transformations mathematically prevents degradation of image quality that would otherwise occur with repeated serial interpolation. Linear transformations can be inverted analytically, and nonlinear transformations can be inverted numerically.

A variety of interpolation models are available, including nearest neighbor, trilinear, sinc, windowed sinc, and chirp z interpolation. Decomposition of affine three-dimensional transformations into a series of three shears (with prepass interpolation to minimize aliasing) allows for particularly efficient implementation of sinc and chirp z interpolation.

Image intensities can be rescaled during intepolation, based on user-defined values or on the scaling factors derived by the minimization algorithm.

Additional utility programs allow for reorientation of images, resizing of images, manual definition of rigid-body transformation parameters, and averaging of images. The default interface uses the command line (on the Macintosh, a window allows entry of command line arguments). On UNIX platforms, a Tcl/Tk graphical user interface for the two most commonly used commands constructs system calls to the command line interface based on menu selections and button presses.

0 0

Post a comment