Robust and efficient 3D registration via depth map-based feature point matching in image-guided neurosurgery

In image-guided neurosurgery, preoperatively acquired diagnostic images (e.g., brain MRI) should be accurately registered to the physical space that is specific to the patient´s intraoperative neuroanatomy. A popular framework of registration requires manual defining corresponding positions of fiducial markers on the patient head and the preoperative brain MRI. The procedure is time-consuming and subjective to intra- and inter-observer variations. Therefore, markerless-based registration becomes increasingly popular. In this paper, we propose an automated markerless registration framework. Instead of using physical markers, we automatically detect feature points in face depth maps. The preoperative facial depth map is extracted from MRI, while the intraoperative map is reconstructed with structured light projection, using phase shifting interferometry. Then, we automatically detect and match the feature points on these two depth maps, using a robust method based on the extended SIFT algorithm. The transform matrix between the two coordinate systems can be computed accordingly. Our experiments on real data result in reasonable registration efficiency, while synthetic testing reveals promising accuracy. Average online processing time is no more than 1s totally in a MATLAB implementation.