Matching strategy for co-registration of surfaces acquired by magnetic resonance imaging
Rita W. T. Cheng, Richard Frayne, Janet Lenore Ronsky, Ayman Fawzy Habib · 2005
Many photogrammetric and GIS applications work with surfaces that are commonly acquired by different sensors, from different viewpoints, and/or at different times. Manipulation of these data requires them to be relative to the same reference frame, and therefore surface matching is a necessary procedure for these applications. Similar to remote sensing, medical image analyses also deal with surfaces such as in studies of disease progression where changes between anatomical surfaces are detected. Magnetic resonance imaging (MRI), a medical imaging modality, is used in this research to capture 3D data of knee joint structures to aid in the non-invasive monitoring of joint diseases. However, human subjects can be positioned differently each time, and disease progression might lead to anatomical changes. As a result, surface matching is an essential task for these applications. Due to the similarities between remote sensing and medical image analyses, the major objective of this research is to translate and modify methods originally developed for geographic data to create new techniques that are feasible for accurate co-registration of MR 3D data. The proposed methodology is based on a voting scheme that can simultaneously establish the correspondence between datasets and estimate the transformation parameters. The matching is performed locally, and only matched features will contribute to the determination of the transformation parameters. Preliminary experiments were conducted on bone surfaces, and an average normal distance between surface elements of 0.201 mm was achieved after registration. This is quite good considering the MR image resolution, and it also shows that the proposed matching strategy is feasible and reliable when applied to MR data.