Ratio of Image Uniformity RIU

The RIU algorithm was originally introduced by Woods et al. [39] for the registration of serial PET studies but has now been applied to serial MR images as well. It is available in the AIR registration package from UCLA. The RIU algorithm finds the transformation that minimizes the standard deviation of the ratio of image intensities. This ratio is computed on a voxel-by-voxel basis from the target image and the transformed source image that results from the current estimate of the registration transformation. The RIU measure is most easily thought of in terms of an intermediate ratio image R comprising N voxels within the overlap domain Q A,B.

r(xa ) = -

qa ,b

qa ,b

R = -1 I R(Xa ) (11) N, xa - qa ,b to be registered. One solution is to transform or re-map the intensities in the image from one modality so that the two images look similar to each other. This has been used with some success by re-mapping high intensities in CT, corresponding to bone, to low intensities [40]. The resulting image has the approximate appearance of an MR image and registration proceeds by maximizing cross correlation. An alternative is to compute differentials of image intensity that should correlate between the two images. Van den Elsen et al. [41] and Maintz et al. [42] compute the edgeness of an image from differential operators applied directly to the image intensities. If the two images have boundaries at corresponding locations then cross correlation of the edgeness measure should be maximal at registration.

Was this article helpful?

0 0

Post a comment