The aim of this project is to provide a clinical evaluation of the accuracy of retrospective techniques for the intermodality registration of volume images of the human head. The goal of retrospective image registration is to provide a non-invasive avenue for the visualization of function (PET) in the context of anatomy (MR) and soft tissue (MR) in the context of bone (CT). Retrospective systems have been under development at many sites for many years, but rigorous evaluation has been limited to phantoms. In this project several retrospective CT-to-MR and PET-to-MR image registration techniques will be applied to images of ten or more patients to determine a rigid transformation to achieve alignment. The accuracy of the transformation will be evaluated for each retrospective technique by comparing it with the transformation determined on the same image sets by a prospective system. The retrospective techniques rely on the localization of anatomical features, while the prospective system is based on the localization of fiducial markers which are implanted in the skull prior to imaging. Because the markers, unlike anatomical features, are tailored to permit localization at a measurably high level of accuracy and because their registration can be effected and checked by deterministic algorithms, the prospective system provides a standard against which the accuracy of the retrospective techniques can be evaluated. The patient images with all traces of the markers removed will be made available at Vanderbilt via electronic transfer to several other sites. Each such site, operating independently, will apply its registration technique to the images and will communicate its transformation parameters back to Vanderbilt for comparison with the standard transformation. The error in positioning relative to the standard will be assessed over the entire visible portion of the brain and also within specific pertinent regions of the brain. To determine the effect of geometric distortion in MR on this error, geometrically corrected MR images will be included with each image set. The accuracy of registrations involving these corrected images will be compared with that for the uncorrected images. The objective assessment provided by this project should help to establish the level of confidence with which neuroradiologists and neurosurgeons can approach retrospective image registration.
Showing the most recent 10 out of 11 publications
