Accurate pet-pet registration to assess lung tumor evolution

Monitoring lung tumors necessitates precise registration of PET images acquired at different stages of a therapy. Existing direct registration methods try to match images leading to distorting the tumor. Inaccurate medical decisions would follow. This paper contributes to solving this problem. The proposed method consists of an indirect registration that takes three different stages. First, multimodal registration of pairs of CT and PET images is performed. Second, CT images are registered to estimate anatomical deformations. Third, the obtained transformation serves for the reconstruction of the PET moving image. The result is a pair of comparable PET images that can be analyzed to assess the actual evolution of the disease. The method has been validated using two types of experiments: phantoms and simulated tumors with synthetic deformations. The paper shows results of these experiments. They prove the validity of the approach and the accuracy of the reconstruction.