Automatic Correction of Registration Errors in Surgical Navigation Systems
Surgical navigation systems are used widely among all fields of modern medicine, including, but not limited to ENT- and maxillofacial surgery. As a fundamental prerequisite for image- guided surgery, intraoperative registration, which maps image to patient coordinates, has been subject to many studies and devel- opments. While registration methods have evolved from invasive procedures like fixed stereotactic frames and implanted fiducial markers toward surface-based registration and noninvasive mark- ers fixed to the patient's skin, even the most sophisticated reg- istration techniques produce an imperfect result. Due to errors introduced during the registration process, the projection of navi- gated instruments into image data deviates up to several millime- ter from the actual position, depending on the applied registration method and the distance between the instrument and the fidu- cial markers. We propose a method that allows to automatically and continually improve registration accuracy during intraopera- tive navigation after the actual registration process has been com- pleted. The projections of navigated instruments into image data are inspected and validated by the navigation software. Errors in image-to-patient registration are identified by calculating intersec- tions between the virtual instruments' axes and surfaces of hard bone tissue extracted from the patient's image data. The infor- mation gained from the identification of such registration errors is then used to improve registration accuracy by adding an additional pair of registration points at every location where an error has been detected. The proposed method was integrated into a surgical nav- igation system based on paired points registration with anatomical landmarks. Experiments were conducted, where registrations with deliberately misplaced point pairs were corrected with automatic error correction. Results showed an improvement in registration quality in all cases.