New biopsy plans are being proposed to improve the detection of significant cancers in the prostate. Among these methods are those that suggest new needle sites, more sites, and also optimal orientations for targeted biopsy. The key challenge being the maximization of the detection of cancers while at the same time using small number of needles to reduce risk of patient anxiety, bleeding and incontinence. To address the fairly consistent spatial inhomogeneity of cancers found among individuals within the population, researchers constructed a cancer atlas of the prostate. The atlas displayed voxel-wise likelihood of cancer occurrences within the prostate, and an optimal 6-7 biopsy scheme to maximize cancer detection. To be useful clinically, the atlas must be warped to the patient's transrectal ultrasound image (TRUS). Previously we suggested use of a shape model to register the surface of the atlas to the segmented prostate surface from the TRUS image reducing the degrees of freedom and computation time. This surface correspondence was used to elastically warp the entire atlas volume to register with the subject. Here we discuss a fast predictive approach where elastic volume warps are also used to train the shape model offline resulting in the construction of an extended shape basis. The surface registration of the atlas and the segmented subject surface yields projections on the extended shape basis from which volume warps can be directly inferred. The accuracy and precision of this predictive method was compared with the original method to analyze the speed-accuracy and precision trade off. We found the average accuracy and precision to be less than 1 mm.