A robust registration method for real-time ultrasound image fusion with pre-acquired 3D dataset

In recent years real-time ultrasound (US) image fusion with pre-acquired 3D dataset has become widely used in both diagnosis and image-guided interventions. The accuracy of a US image fusion system heavily depends on the image registration method. However, the registration procedure of this application is inevitably interfered by possible outliers in the corresponding point pairs. This is either caused by image feature difference between two modalities or by tissue shifting and deformation of patient body between two imaging studies. While traditional methods often ignore the position error of registration points, we present a random sample consensus-based algorithm to reduce the impact of outliers and improve the robustness. To evaluate our algorithm, a simulation study is carried out, and the new method is compared with state-of-the-art, least square (LS) method. It is shown that our new method is comparable with LS method under non-outlier condition, but it performs significantly better when outliers exist.