Registration in neurosurgery and neuroradiotherapy applications

Because of the high level of accuracy needed in neurosurgery, many computer-assisted surgery (CAS) and augmented reality techniques have been developed in this field. A common issue with all of these techniques is registration between preoperative three-dimensional images (computed tomography and magnetic resonance imaging) and the patient in the operating room. We present, in the first part of this paper, a survey of the latest CAS technologies, using fully automatic registration without fiducial landmarks. All of the registration algorithms described are based on minimization of a cost function. We then describe our approach. Our cost function is simply the mean square error (MSE), minimized by the iterative closest point algorithm (ICP). Because the weak point of the ICP algorithm is the closest point computational cost, we precalculate it by a "closest point map," inspired from classical distance map. We finally perturb the found solution to eliminate local minima close to the global minimum. This paper summarizes the various methods presented. We study the shape of the different cost functions and show that there is no need for a complex cost function. MSE has sufficiently good convergence properties to reach a position very close to the global minimum. We also demonstrate the influence of a final perturbation of the found solution to improve registration. Finally, we test the registration on different regions of the patient's head.