An efficient feature-based non-rigid registration of multiphase liver CT images using matching region partition

Image registrationis a preliminary component in medical image analysis, which can significantly improve radiologists' performance in detecting and characterizing the lesion. In this paper, we propose an efficient feature-based non-rigid registration approach for multiphase liver CT images. The proposed method begins with extracting corners and edges simultaneously from reference and floating images using Harris corner detector. Then joint descriptor combing local descriptor (SIFT) with global corresponding notion (Shape Context) is applied to describe those features. In matching stage, we propose to distinguish liver area with the rest of abdominal region by partitioning the entire abdominal region into two sub regions, i.e., inner matching region and outer matching region, by an automatic segmentation approach. Principally, our matching is independently performed on these two regions, which makes matching more reliable and accurate. Eventually, Thin-plate splines scheme is used to model deformation field between correspondences. An evaluation of the algorithm on clinical data sets is presented to validate its efficiency.