Comparison of an image registration technique based on normalized mutual information with a standard method utilizing implanted markers in the staged radiosurgical treatment of large arteriovenous malformations.
Academic Article
Overview
abstract
PURPOSE: To compare a noninvasive technique based on normalized mutual information for registering image sets associated with staged radiosurgical treatments of large arteriovenous malformations (AVMs), with a gold-standard method using radiographically evident markers implanted in the skull. METHODS: Nine patients receiving multistage treatment of large AVMs at the University of California at San Francisco (UCSF) gamma knife facility were included in this study. For each patient, the transformations of shot coordinates between a reference treatment stage and subsequent treatment stages were determined at UCSF, based on radiographically defined coordinates of implanted markers in each stereotactic space. A magnetic resonance (MR) image set was acquired for each treatment stage, and used for treatment planning. The two MR image sets for each treatment pair were sent to Yale for an unbiased, independent analysis of shot transformations. An image registration technique based on normalized mutual information was used to produce a single fused image study for each treatment pair. External copper sulfate fiducial markers for both image sets were evident on the fused images, allowing coordinates in both stereotactic systems to be defined. Coordinate transformation between the two systems was determined, based on digitized coordinates of seven common fiducial marker images. RESULTS: The average measured overall root-mean-square discrepancy between the Yale and UCSF transformed shot coordinates is 1.1 +/- 0.3 mm. The corresponding error in Yale transformed coordinates is 1.0 +/- 0.3 mm, assuming an inherent 0.5 mm error in the UCSF method. CONCLUSIONS: The normalized mutual information method can be used to obtain good image registration between successive sessions in staged treatments. Further improvements in the reported methodology are outlined. Because the mutual information method is less invasive than the implanted marker method, it may be preferable in many cases.
