In order to improve the accuracy in thyroid tumor detection, a registration of SPECT image and B-type ultrasound image based on artificial bee colony algorithm is improved. First, NSCT-GCBAC segmentation and KFCM segmentation are used respectively to extract the contours, according to the different characteristics of thyroid SPECT image and B-type ultrasound image from two imaging devices. Second, mutual information is used for similarity measure, and affine transform model is adopted to solve the space transform parameters. Finally, an improved artificial bee colony algorithm is used to optimize the space registration transform parameters. The experimental results show that the method can effectively improve the speed and the effect of registration.