Registration of Non-Segmented Images Using a Genetic Algorithm

The proposed method aims at solving the global 3—D automatic non—rigid registration problem of two volume datasets coming from the same modality without any segmentation. The tested algorithm combines a voxel based approach with an optimization procedure carried out by a simple genetic algorithm. The method is based on the search for the polynomial warping model which minimizes a L 1 distance between two volumes datasets. To avoid prohibitive computational cost, the genetic algorithm uses a stochastic fitness function which operates on randomly selected measure sites in neighbourhoods of contours, with an adaptive search space scaling scheme. Application of the method is made to the elastic registration of 3—D CT volumes and gives good results in a reasonable processing time.