Automated rejection of contaminated surface measurements for improved surface registration in image guided neurosurgery.

Most image guided Neurosurgery employs adhesively mounted external fiducials for registration of medical images to the surgical workspace. Due to high logistical costs associated with these artificial landmarks, we strive to eliminate the need for these markers. At our institution, we developed a handheld laser stripe triangulation device to capture the surface contours of the patient's head while oriented for surgery. Anatomical surface registration algorithms rely on the assumption that the patient's anatomy bears the same geometry as the 3D model of the patient constructed from the imaging modality employed. During the time interval from which the patient is imaged and placed in the Mayfield head clamp in the operating room, the skin of the head bulges at the pinsite and the skull fixation equipment itself optically interferes with the image capture laser. We have developed software to reject points belonging to objects of known geometry while calculating the registration. During the course of development of the laser scanning unit, we have acquired surface contours of 13 patients and 2 cadavers. Initial analysis revealed that this automated rejection of points improved the registrations in all cases, but the accuracy of the fiducial method was not surpassed. Only points belonging to the offending instrument are removed. Skin bulges caused by the clamps and instruments remain in the data. We anticipate that careful removal of the points in these skin bulges will yield registrations that at least match the accuracy of the fiducial method.