Automated prostate localization in external-beam radiotherapy using mutual information-based registration of treatment planning CT and daily 3D ultrasound images

The accuracy of mutual information-based registration for automatic alignment of planning CT and daily 3D ultrasound images is evaluated. The results show the importance of preprocessing both the CT and the daily 3D ultrasound images to maximize registration accuracy. Contours traced by experts for treatment planning were used as a priori information for contrast enhancement in the CT images. Furthermore, the CT and daily 3D ultrasound datasets were filtered using edge preserving anisotropic diffusion filtering. The study also compared the accuracy of rigid registration with three or six degrees of freedom against elastic registration based on a deformation field modelled using B-Splines. Experiments with clinical images of wide ranging image quality acquired using a SonArray™ system achieved success ratios in estimating the prostate shifts of 80%, 60% and 93% for rigid registration with three or six degrees of freedom and elastic registration respectively.