We have developed a general purpose registration algorithm for medical images/volumes. Given a source and target image we automatically estimate a smooth warp field that brings the source image into register with the target image.

We model the transformation between images as locally affine but globally smooth. The model also explicitly accounts for local and global variations in image intensities, and for missing data between the source and target. This approach is built upon a differential multiscale framework, allowing us to capture both large and small scale transformations.

Example registration results are shown to the right. This animated gif shows a 3-D example - on the left is the source (with more than 1/2 of the head missing), in the middle is the target, and on the right is the registered source (with the original head spliced in to show the accuracy of the registration).

Matlab code (qr_0.1.tar.gz)
Matlab code (qr_0.2.tar.gz)
(version 0.2 incorporates the EM algorithm to contend with missing data)

(Collaborative work with Senthil Periaswamy)