Augmented reality visualization system for intravascular neurosurgery

We aimed to construct an augmented reality-based visualization system to support intravascular neurosurgery and evaluate it in clinical environments. Three-dimensional (3D) vascular models are overlaid on motion pictures from X-ray fluoroscopy by 2D/3D registration using fiducial markers. The models are reconstructed from 3D data obtained from X-ray computed tomographic angiography or from magnetic resonance angiography using the marching-cube algorithm. Intraoperative X-ray images are mapped as texture patterns on a screen object which is displayed with the vascular models. Distortion of X-ray fluoroscopy is eliminated by a new technique of screen mesh deformation. A quantity called reprojection distance was introduced to evaluate the reliability of the displayed images. It predicts the maximum registration error around the registered objects. Analyses of reprojection distances were performed using synthetic data consisting of marker coordinates with 2D or 3D errors. The tolerance of reprojection distance for the clinical environment was determined to be 3.0 mm. The system was tested in two clinical cases in which reprojection distances of 2.6 and 2.09 mm were obtained. Construction and evaluation of our prototype system were successfully carried out. Further development is planned employing a range sensor to permit markerless registration.