Surface-based registration of CT images to physical space for image-guided surgery of the spine

The objective of this work is to investigate the registration of physical vertebral points to triangulated surfaces extracted from computed tomography images. Ultimately, this research is intended to advance the larger goal of developing an image-guided surgical technique for open- or closed-back surgery. The use of image-guided techniques will ideally result in surgical procedures that are more accurate and less invasive than traditional methods.; Performance of image-guided surgery requires accurate intra-operative registration, which in turn requires accurate pre-operative segmentation. A clinically viable combination of automatic vertebral segmentation and intra-operative registration has not yet been reported in the literature. We propose using a modified version of the classical marching cubes algorithm for segmentation and a surface-based registration algorithm for intra-operative registration. We also suggest that point collection be limited to easily accessible portions of the vertebral surface, and we provide quantitative measurements to describe how the use of only these selected surface areas affects registration accuracy.; We have conducted numerous experiments using two plastic phantoms, six patients, and all five lumbar vertebrae. Our results show that accurate registration between image space and physical space can be achieved using automatically extracted surfaces and points that represent a small fraction of the vertebral surface. Specifically, we have examined the effects of using variations of surface extraction and simplification methods within the general class of iso-intensity-based algorithms. We have also examined registration accuracy achievable using only the tip of the spinous process and the central laminar regions, since these portions of the vertebral surface are most likely to be accessible during either open-back surgery (via a tracked probe) or closed-back surgery (via ultrasound imaging). We have investigated several additional parameters that affect surface extraction and point set collection, and we have determined in every experimental case that a point set comprising the spinous process tip and the central laminar regions can be registered to automatically extracted surfaces with sufficient accuracy. In most cases, registration errors measured at clinically viable targets are sub-millimetric.