A Comparison Of US Versus MR Based 3 D Prostate Shapes Using Radial Basis Function Interpolation And Statistical Shape Models

This paper presents a comparison of 3D segmentations of the prostate, based on 2D manually-segmented contours, obtained using ultrasound (US) and magnetic resonance (MR) imaging data collected from 40 patients diagnosed with localized prostate cancer and scheduled to receive brachytherapy treatment. The approach we propose here for 3D prostate segmentation first uses radial basis function interpolation to construct a 3D point distribution model for each prostate. Next, a modified principal axis transformation is utilized for rigid registration of the US and MR images of the same prostate in preparation for the following shape comparison. Then, statistical shape models are used to capture the segmented 3D prostate geometries for the subsequent cross-modality comparison. Our study includes not only cross-modality geometric comparisons in terms of prostate volumes and dimensions, but also an investigation of interchangeability of the two imaging modalities in terms of automatic contour segmentation at the pre-implant planning stage of prostate brachytherapy treatment. By developing a new scheme to compare the two imaging modalities in terms of the segmented 3D shapes, we have taken a first step necessary for building coupled US-MR segmentation strategies for prostate brachytherapy pre-implant planning, which at present is predominantly informed by ultrasound images only.