Optimization and Quantification of Aaccuracy for Rigid Point Based Registration for Computer Aided Surgery

Abstract

When conducting complex neurosurgical interventions on the brain, surgeons are often assisted by planning systems and neuronavigation. To position the instruments in relation to the preoperatively acquired image datasets, rigid registration between the coordination systems of the patient and the corresponding diagnostic images is performed using fiducial markers or screws attached to the patient's head. In this work a registration-framework was implemented, which allows the easy integration of different point based registration algorithms as well as the estimation of registration accuracy, To perform the registration, two algorithms were implemented and tested using simulation as well as real world measurements. The target registration error (TRE) is predicted by a method introduced from Fitzpatrick. Two different approaches were implemented and both achieve better results than the RMS Error ({''}Root Mean Square{''}) widely used in clinical applications. To provide the surgeon with an intuitive understanding of the accuracy, a graphical representation of the estimator was developed and integrated into the existing MOPS 3D planning system. Additionally an iterative marker selection method was developed for identifying markers which would contribute to a higher registration error. The described registration framework is developed in C++ using MS Visual Studio and is easy to integrate in different medical applications.