2D/3D Registration for Localization of Mammographically Depicted Lesions in Breast MRI

X-ray mammography (XRM) and Magnetic Resonance Imaging (MRI) are likely to provide complementary diagnostic information for early breast cancer detection. However, topographic correlation of both modalities is challenging due to different dimensionality of images, patient positioning and compression state of the breast. In this paper we present an automated registration method, which allows prediction of the position of a lesion in the contrary modality. It is based on a FEM simulation mimicking the mammographic compression and is carried out using a patient-specific biomechanical model. An intensity-based optimization of the registration parameters is proposed to incorporate with the clinical variability of datasets. After registration, the position of a point of interest can be estimated within the three-dimensional MRI volume based on two mammograms acquired from different projection angles. The method was evaluated with 47 datasets from clinical routine. The mean registration error for localizing a lesion in the 3D MRI volume was 14.3 mm. The automatic registration method enables localization of e.g. microcalcifications which are only visible in XRM, within the corresponding MRI volume. It is therefore likely to assist radiologists in multimodal diagnosis.