Automatic registration of FDG_CT and FLT_CT images integrating Genetic Algorithm, Powell method and wavelet decomposition

This paper describes a novel mutual information-based registration method that integrates the use of a Genetic Algorithm (GA), the Powell method (PM), and Wavelet decomposition in order to register in an optimal fashion the fluorodeoxyglucose (FDG)_CT and fluorodeoxythymidine (FLT)_CT image modalities. By registering these two computed tomography (CT) modalities, we combine the strengths of the two radiotracers knowing that FDG uptake is higher in cancerous lesions, while FLT uptake is closely correlated with cellular proliferation. Registration through these tracers, FDG and FLT, increase both sensitivity and specificity for imaging cancer, and is essential for optimizing the results of the diagnosis. In this study, this integrated approach, we refer to as GPW, focuses on solving three problems: (1) Reducing the computational time of GA when it is searching for the best global solution; (2) Preventing the PM method to fall into a local solution for image registration; (3) Providing the necessary image pre-processing steps for enhanced feature analysis of FDG_CT and FLT_CT images. After registration, the location of the cancerous lesions on the liver could be observed directly on the FLT_CT image. When registering wavelet decomposition images, the GA is applied for determining the maximal value of the normalized mutual information between a reference image and a moving image. The Powell method (PM) is implemented in search for the best solution starting from an initial set of registration points.