Registration of brain images by surface matching. A CT to CT validation

Retrospective registration of CT brain images allows better utilization of their complementary information in medical diagnosis, surgery planning and medical treatment. We present a CT-CT validation of a new registration method using phantom and patient data acquired by a helical scanner. The method involves three steps. First, images of the surfaces (bone or skin) of the two volumes are generated in several orientations using semi-automated thresholding and raycasting techniques. Next, the operator selects the regions of interest (ROIs) on the surface of the first volume that are to be matched with the surface of the second volume. Finally, an iterative algorithm based on the downhill simplex optimization method is used to find the spatial relationship that best fits the ROIs onto the surface of the second volume. Validation studies conducted on CT phantom data revealed the sensitivity of the method to the size and surface relief of the ROIs. A sub-millimeter accuracy was easily achieved with a mean value of 0.29 mm and a reliability of 91.8%, by using three ROIs for the matching. The accuracy of the method was confirmed visually on patient data. (author)