<title>Comparison of three methods for registration of abdominal/pelvic volume data sets from functional-anatomic scans</title>

Abstract

The purpose of this work was to evaluate three volumetric registration methods in terms of technique, user-friendliness and time requirements. CT and SPECT data from 1 1 patients were interactively registered using: a 3D method involving only affmne transformation; a mixed 3D-2D non-affine (warping) method; and a 3D non-affine (warping) method. In the first method representative isosurfaces are generated from the anatomical images. Registration proceeds through translation, rotation, and scaling in all three space variables. Resulting isosurfaces are fused and quantitative measurements are possible. In the second method, the 3D volumes are rendered co-planar by performing a oblique projection. Corresponding landmark pairs are chosen on matching axial slice sets. A polynomial warp is then applied. This method has undergone extensive validation and was used to evaluate the results. The third method employs visualization tools. The data model allows images to be localized within two separate volumes. Landmarks are chosen on separate slices. Polynomial warping coefficients are generated and data points from one volume are moved to the corresponding new positions. The two landmark methods were the least time consuming (1 0 to 30 minutes from start to finish), but did demand a good knowledge of anatomy. The affine method was tedious and required a fair understanding of 3D geometry.