The aim of this project is to provide an interactive, World Wide Web based interface that may be used by external researchers to evaluate the accuracy of their retrospective image registration techniques on volume images of the human head. Retrospective registration provides 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). In this project CT, MR, and PET volumes from over one hundred patients will be made available to qualified researchers. For some image volumes, the researchers will perform retrospective registrations on the image volumes and input the results into the interface on the Web page maintained by Vanderbilt. Programs embedded in the Web page will compare the retrospective results with the results of a prospective system, and return an accuracy estimate for the retrospective technique. For the rest of the image volumes, researchers will submit their registration results to Vanderbilt by e-mail, and receive error estimates from Vanderbilt by e-mail: the researchers thus stay blind to the gold standard transformations used for evaluation. The retrospective techniques rely on anatomical features, while the prospective system uses fiducial markers that are implanted in the skull. Because the markers permit localization at a measurably high level of accuracy and because they can be registered with deterministic algorithms, the prospective system provides a gold standard of accuracy. Because of recent advances made at Vanderbilt in predicting the error of the gold standard system, this error may be taken into account when the retrospective techniques are evaluated. The Web site for this project will contain two sections: a public section, freely accessible and containing information and links to pertinent publications concerning the project, and a private section, accessible only by a login and password. This login and password will be given to researchers who request access to the images and satisfy Vanderbilt's IRB that they will treat the images properly. On the private section of the site, all the image volumes will be available with all traces of the fiducial markers removed, in addition to an interface allowing the researchers to evaluate their registration results. Approximately half of the available image volumes will be provided for this anonymous, interactive evaluation protocol, along with the gold standard transformations for these volumes; the rest of the images will be retained for the traditional, blinded evaluation protocol that we have been using for several years. The objective assessment provided by this project should foster improvement of the accuracy of retrospective registration techniques.
| Dogdas, Belma; Shattuck, David W; Leahy, Richard M (2005) Segmentation of skull and scalp in 3-D human MRI using mathematical morphology. Hum Brain Mapp 26:273-85 |
| Ren, Hai-Ping; Yang, Hu; Ma, Bing-Rong et al. (2003) A novel magnetic resonance-positive emission image registration based on morphology. Biomed Mater Eng 13:187-96 |
