OBJECTIVETo test the spatial accuracy of coordinates generated from magnetic resonance imaging (MRI) scans, using the Brown-Roberts-Wells head frame and localizer system (Radionics, Inc., Burlington, MA).METHODSAn anthropomorphic head phantom, consisting of a two-dimensional lattice of acrylic spheres (4-mm diameter) spaced 10 mm apart and embedded in a brain tissue-mimicking gelatin-agar gel, was constructed. The intersphere distances for the target lattice positions in MRI and computed tomographic scan sets were compared. The data sets were fused, and differences in fiducial marker and intraphantom target positions were measured.RESULTSIntersphere distances were identical for the MRI and computed tomographic scan sets (10 ± 0.1 mm). Differences in fiducial marker positions [maximal lateral difference, 0.97 mm; mean absolute lateral difference, 0.69 ± 0.22 mm; maximal anteroposterior (AP) difference, 1.99 mm; mean absolute AP difference, 1.29 ± 0.67 mm] were correlated with differences in intraphantom target positions (maximal lateral difference, 0.83 mm; mean absolute lateral difference, 0.28 ± 0.24 mm; maximal AP difference, −1.97 mm; mean absolute AP difference, 1.63 ± 25 mm; maximal vertical difference, −0.73 mm; mean absolute vertical difference, 0.34 ± 0.21 mm). This suggested that improper fiducial rod identification and the subsequent transformation to stereotactic coordinate space were the greatest sources of spatial uncertainty.CONCLUSIONWith computed tomographic data as the standard, these differences resulted in maximal and minimal composite uncertainties of 2.06 and 1.17 mm, respectively. The measured uncertainties exceed recommended standards for radiosurgery but allow the possible use of MRI-based stereotactic treatment planning for certain intracranial lesions, if the errors are corrected using appropriate software. Clinicians must recognize that error magnitudes vary for different systems, and they should perform systematic, scheduled, institutional error analyses as part of their ongoing quality assurance processes. This phantom provides one tool for measuring such variances.