Fostering innovations, nurturing health.
An internet-connected, patient-specific, deformable brain atlas integrated into a surgical navigation system
Armond L. Levy, Timothy J. Schaewe, Michael I. Miller, Kurt R. Smith, Abed M. Hammoud, Jaimie M. Henderson, Sarang Joshi, Kevin E. Mark, Christopher D. Sturm, Leslie L. McDurmont, Jr. and Richard D. Bucholz
Journal of Digital Imaging August 1997 - Standard anatomic information, such as that contained in atlases, is usually based upon individual human specimens, and therefore it can- not account for structural variability between patients. Thus, utilizing atlas information regarding specific locations and dimensions of anatomic structures could be misleading or incorrect if applied, without complex mapping criteria, to specific patients. This shortcoming is particularly noteworthy in the case of neuroanatomy, where brain structures may be very small, and yet may vary proportionally greatly in size, location, or even existence.
The advent of the World Wide Web (WWW) has already begun to affect the practice of medicine, especially the technologically oriented field of neurosurgery. Incorporating the capabilities of the WWW into the operating room setting could permit such advances as intraoperative enrollment into multi-center protocols, remote consultation, and demonstration of surgical techniques.l Solutions to the need for patient-specific anatomic information could be greatly enhanced by the advent of informational sources on the WWW. Mapping of patient- specific anatomic structures, pertinent to a particular surgical procedure, to corresponding WWW pages, with real-time access in the operating room, is a powerful new application of this technologic capability.
We are developing a three-dimensional brain atlas that integrates seamlessly with our existing frameless stereotactic, surgical navigation system. This new atlas will address anatomic variability by utilizing ah image deformation algorithm to tailor standard atlas views to each individual patient's radiologic brain studies. This system will enable several distinct advantages over current technology, including "virtual" visualization of patient brain structures as high-resolution photographic images, real-time stereolocation of the surgical act within this three-dimensional virtual image, and integration via the Internet of neuroscience knowledge with surgery (see ref. 1 for ah extensive list of neurosurgically related WWW sites).
Surgical Navigation System
We have previously developed and employed clinically an optically based frameless surgical navigation system (StealthStation, Surgical Naviga- tion Technologies, Boulder, CO) marketed by So- famor Danek Group of Memphis, TN. 2,3 The sys- tem consists of a UNIX-based computer workstation (Silicon Graphics Inc, Mountain View, CA) and an infrared optical digitizer (Flashpoint 5000, Image Guided Technologies, Boulder, CO) to display position real-time on a high-resolution monitor. The digitizer uses a charge-coupled camera array suspended from the operating room ceiling to track light-emitting diodes attached to a rigid reference array mounted to a Mayfield head clamp. The position of surgical instruments, modified by the addition of light-emitting diodes, is determined in the coordinate system of the reference array.
An accurate simple overlay of atlas images onto patient radiographs is impossible because of inter- patient geometric anatomic variability. 4 Therefore, the atlas image must be "deformed" to match each individual. A hierarchic algorithm unifying both landmark-based and intensity-based transforma- tions was developed to address this issue, s,6 The algorithm creates iterative solutions of a system of elasticity-based partial differential equations (PDE), which are applied as vector fields to the coordinate system of the template (atlas) image.
Landmarks and linear manifolds such as sulci are identified in the dataset to serve as manifolds by which the template and target (patient) images are coarsely registered. Next, the full volumetric data are addressed. Both the difference between template and target intensities, and a constraint main- taining the relative structural topology of the tem- plate volume, are considered. This constraint is modeled by PDEs derived by modeling elastic deformations. The final transformation is accomplished via a course-to-fine iterative solution of the model-based PDEs on the full volume.
Our atlas is based upon the Visible Human dataset. Brain structures of zero, one, two, and three dimensions are segmented onto the three- dimensional photographic template. A "segmenta- tion map" file results. This file is then accessed asa look-up table from whicha unique brain structure results from each voxel in the dataset, permitting mutually exclusive, real-time linkage to the WWW site(s) relevant to the selected brain structure. Furthermore, this segmentation map file can un- dergo deformation along with its corresponding template images by application of the identical vector fields described previously.
All images are handled in standard Analyze (Biomedical Imaging Resource, Mayo Foundation, Rochester, MN) format, and they are imported directly into the surgical navigation system, which has been enhanced to handle the display of two complete volumetric datasets. The deformed photo- graphic arias, carrying its full segmentation, can be superimposed upon the patient dataset in real-time, and ir can be shaded in or out by variable blending. The navigational system is therefore able to not only communicate the position of the surgical instrument with respect to preoperative images but also to determine which brain structure is at the tip of the surgical instrument based on the segmented dataset.
The final element of this "living atlas" is the linkage of each segmented anatomic feature to a page in the WWW. As the segmentation map is deformed along with the cryostatic template, the WWW Iook-up table can also be effectively de- formed. By embedding a browser witbin the surgi- cal navigation system, the current location of the instmment is correlated with a Universal Resource Locator (URL) on the Internet, which can be displayed for review. For example, if the neurosur- geon "clicks" while an instrument is within the hippocampus, that structure is highlighted on the superimposed deformed photograph and the pa- tient's three-dimensionally reconstructed MR im- age; at the same time, the appropriate hippocampus URL is retrieved from the WWW. This brings a wealth of current neuroscience information into the operating room in a fashion relevant to the hippocampus, of to any addressed brain structure(s) in real-time.
DISCUSSION AND CONCLUSIONS
...We present a patient-specific deformable atlas that is easily integrated into our surgical navigation device. The atlas is a powerful new tool for the integration of information with the surgical act. This is the first time an outline atlas has been implemented to provide individualized, high- resolution data to the operating surgeon (see refs. 13 and 14). Our basic paradigm allows surgical navigation based upon photographic images, with a resultant great improvement in tissue contrast, and hence discernibility of small brain structures, over the best MR images. As such, this tool will allow preoperative and intraoperative guidance that is superior to any current system. It may, for example, improve preoperative selection of pallidotomy targets, thus eliminating the need for expensive and time-consuming intraoperative location procedures such as ventriculography. In addition, the introduction of the vast resources of neuroscience into the operating room via the WWW could revolutionize intraoperative decision-making and techniques.