Non-rigid registration of tomographic images with Fourier transforms

A.R. OSORIO; R. A. ISOARDI; G. MATO

Journal of Physics Conference Series

IOP Publishing Ltd.

Año: 2007 vol. 90 p. 1 - 9

1742-6588

Spatial image registration of deformable body parts such as thorax and abdomen has important medical applications, but at the same time, it represents an important computational challenge. In this work we propose an automatic algorithm to perform non-rigid registration of tomographic images using a non-rigid model based on Fourier transforms. As a measure of similarity we use the correlation coefficient, finding that the optimal order of the transformation is n=3 (36 parameters). We apply this method to a digital phantom and to 7 pairs of patient images corresponding to clinical CT scans. The preliminary results indicate a fairly good agreement according to medical experts, with an average registration error of 2 mm for the case of clinical images. For 2D images (dimensions 512×512), the average running time for the algorithm is 15 seconds using a standard personal computer. Summarizing, we find that intra-modality registration of the abdomen can be achieved with acceptable accuracy for slight deformations and can be extended to 3D with a reasonable execution time.