Registration and statistical analysis of PET images using the wavelet transform

We have described a general procedure for the processing and analysis of PET data. We have used the multiresolution framework of the wavelet transform to derive new solutions for the two main processing steps. The first task was to align the various brain images using a general affine deformation model. Our registration procedure uses a continuous polynomial spline image model and takes advantage of the multiresolution structure of the underlying function spaces. This method implements a nonlinear least squares optimization technique with a coarse-to-fine iteration strategy that substantially improves the overall performance of the algorithm. The second task was to analyze the series of registered images and to detect the between group differences in metabolic brain activity. We chose to take advantage of the orthogonality and localization properties of the wavelet transform. Our approach was to apply this transform to the group-difference image and identify the wavelet channels that are globally significantly different from noise