Motivation: Needle-based biopsy and local therapy of prostate cancer depend multimodal
imaging for both target planning and needle guidance. The clinical process involves
selection of target locations in a pre-operative image volume and registering these
to an intra-operative volume. Registration inaccuracies inevitably lead to targeting
error, a major clinical concern. The analysis of targeting error requires a large
number of images with known ground truth, which has been infeasible even for the largest
research centers. Methods: We propose to generate realistic prostate imaging data
in a controllable way, with known ground truth, by simulation of prostate size, shape,
motion and deformation typically encountered in prostatic needle placement. This data
is then used to evaluate a given registration algorithm, by testing its ability to
reproduce ground truth contours, motions and deformations. The method builds on statistical
shape atlas to generate large number of realistic prostate shapes and finite element
modeling to generate high-fidelity deformations, while segmentation error is simulated
by warping the ground truth data in specific prostate regions. Expected target registration
error (TRE) is computed as a vector field. Results: The simulator was configured to
evaluate the TRE when using a surface-based rigid registration algorithm in a typical
prostate biopsy targeting scenario. Simulator parameters, such as segmentation error
and deformation, were determined by measurements in clinical images. Turnaround time
for the full simulation of one test case was below 3 minutes. The simulator is customizable
for testing, comparing, optimizing segmentation and registration methods and is independent
of the imaging modalities used.
