Image registration of MR and PET based on surface matching and principal axes fitting

We present a surface based image registration algorithm, which uses a combination of minimization of average point-to-surface distance and iterative principal axes fitting (IPAF). MR and PET brain contours are automatically selected from their intensity histograms, such that the separability of object and background is maximized. Cubic B-spline surface is introduced to represent brain surface with few parameters. Its use allows us to accurately compute the point-to-surface distance and the gradient of the distance with respect to registration parameters. As a result, effective distance minimization could be achieved by using a gradient descent algorithm if it is not trapped at a poor local minimum. To prevent being trapped, IPAF algorithm is developed to provide a good initial solution. In addition, it greatly speeds up the registration procedure. The entire procedure is fully automated and is fast enough for routine clinical use. Its performance has been tested using 3D phantom and patient data