In clinical diagnosis, and planning and evaluation of therapy, are often supported by several imaging modalities [16]. Different modalities usually provide complementary information. For example, CT (computed tomography) images provide anatomical information of human body in radiotherapy. PET (positron emission tomography) image provide the activity of metabolism of organs, which helps clinician to realize the biochemistry and physiology of the human body in radiotherapy [15]. Due to the PET image being functional but low-resolution, to segment the geometrical feature from the images is difficult, and the registration
process might be failure. Using Mutual Information (MI) [4] of PET and CT images to register both images, clinicians can get more information on focal part. The multi-modality registration of processing provides more information than single-modality.But the computing time to find the maximization of MI is expansive, and maybe falls down the local maximum.To improve the drawback of MI, in the study, we propose an improved algorithm shown as Figure 1. First, we use ideal Mid-Sagittal Plane (iMSP) Algorithm [6] and Z-axis shifting to align CT and PET images in the center position and same slice location in rough. Second, we calculate the MI of different image slices, to find the maximum MI as aligned exactly.
