SIFT and homographic transform-based cardiac image registration

Abstract

Registration of images, serving as a pre-cursor to many medical image analysis methods such as fusion and object detection, remains a challenging task. In this paper, we demonstrate a novel automated mono-modal medical image registration framework of the human heart. Scale Invariant Feature Transform (SIFT) is utilized to map both the reference and floating images, and the homographic transform subsequently performs the necessary alignment. A range of histogram comparison methods then evaluates the results of our registration. Overall, the positive performance shown in our results affirms the potential use of this method for more advanced registration applications in the future, thereby further assisting the medical practitioner in cardiac diagnosis and surgery.