Clinical diagnoses, therapy planning and evaluation, are increasingly supported by images generated by more than one imaging device. To combine the usually complementary information provided by the individual images, they need to be matched. Although many methods have been described to match brain images, few methods have been applied outside the brain. Methods One rigid transformation suffices to describe the match between two brain images. Since the spine articulates, the transformation between two spine images consists of a more complex transformation. Each single vertebral body is a rigid bone entity that can be mapped onto another image using a rigid transformation. Our approach is to split the spine CT volume into subvolumes comprising individual bone segments, and to compute a separate transformation for each subvolume. User input is restricted to indicating the corresponding vertebra in MR for each vertebral subvolume in CT. Our method involves intensity mapping of the CT image to extract approximately all the bone information. Threedimensional cross correlation techniques are used to determine the best mapping from the coordinate system of the mapped CT subvolumes into the original MR volume. Discussion and Results Four sets of spine scans acquired with various imaging protocols were matched using the described method described above, totaling fifteen vertebra matches. In each patient, the accuracy of all matching results was assessed visually in three orthogonal directions. Our preliminary results indicate high accuracy with low sensitivity to the noise level in the images. Our method for matching CT and MR volumes of the spine does not require an external fiducial system to be mounted on the patient. Although a minimal amount of user interaction is required, the results are devoid of user subjectivity. Results on four patients indicate high-accuracy matching with low sensitivity for noise. Development of objective methods for validation of spine matching results is in progress.
Showing the most recent 10 out of 446 publications
