Fiducial-based registration with a touchable region model
Although frameless stereotactic neuro-surgical navigation systems are widely used in many neuro-surgical centers around the world, most of the systems still require the user to define the position of fiducial markers manually from patient scans, a procedure that is tedious, time consuming and often inaccurate. Some researchers have addressed this problem, but they acknowledge that their 2D image processing approach has limitation. We propose a new automatic approach for 3D localization of the fiducial markers, which provides higher 3D localization accuracy, and is independent of the geometry of the marker. Our approach includes three steps. First, sets of 3D morphological operations are employed to extract the candidate fiducial markers as the “seeds”. Then a “conditional dilation” technique is employed to reconstruct the regions of fiducials from the “seeds” which are sifted by several knowledge-based rules. Lastly, the intensity-weighted centroid of each extracted fiducial region is calculated as our final fiducial position. The approach is validated by simulated datasets and a CT phantom scan where the average Fiducial Localization Error (FLE) is 0.37mm and 0.31mm, respectively.