Propagation of the MRI prostate delineation to the planning CT: A new matching contour framework

Abstract

Although accurate delineations of the prostate on computed tomography (CT) images are required for the planning in prostate cancer radiotherapy, these images do not provide the reliable soft tissue contrast. On the contrary, magnetic resonance (MR) images offer the possibility to better delineate both the tumor and healthy prostate tissues. Because CT is still used during the planning, MRICT image registration is an essential step to improve the targeting. In this paper, we propose a new framework to propagate the MRI prostate delineation to the CT-scan based on a robust contour matching approach. Prostate boundaries in CT are characterized with several multi-scale features and detected with a support vector machine (SVM) classifier. A new cost function for aligning the MRI delineation to the detected contours was developed. We evaluated the proposed approach on 11 manually aligned and delineated MR and CT images. The method outperformed the widely used mutual information (MI) and demonstrated the drawbacks of this metric for this application.