Fully automated establishment of stereotaxic image orientation in six degrees of freedom for technetium-99m-ECD brain SPECT.

UNLABELLED Anatomical localization requires establishing an anatomical space within the image matrix. We developed a fast, fully automated method to establish the image orientation for 99mTc-ethylcysteinate dimer (ECD) brain SPECT images. METHODS The image orientation of ECD brain SPECT images was established in four stages. First, the brain surface was edge-detected as an isosurface at an adaptive threshold. Second, a "convex hull" was determined for the isosurface to minimize regional variability in brain shape. A principal axis transformation and a symmetry vector analysis were applied to the convex hull to resolve the craniocaudal direction and to estimate the midsagittal plane. Third, the brain orientation was refined from this estimate by location of the interhemispheric fissure, the tentorial groove and the frontotemporal groove on the isosurface. Last, the intercommissural (anterior commissure-posterior commissure, or AC-PC) line was detected on the midsagittal slice, and the Talairach grid was scaled to fit the maximal brain dimensions from the AC-PC line. RESULTS The average absolute errors were 2.3 degrees +/- 1.5 degrees and 1.08 mm +/- 1.11 mm for the midsagittal plane (n = 24) and 2.04 degrees +/- 0.80 degrees, 2.0% +/- 1.8% of the brain length and 2.3% +/- 2.2% of the brain height for the AC-PC line (n = 8). In addition, this program successfully established the image orientation in 94 of 100 clinical ECD brain SPECT studies. Processing time was <40 sec for 128 x 128 x 50 matrices on a DEC Alpha workstation. CONCLUSION We have developed a fast, robust and fully automated method that determines the orientation of ECD brain SPECT images. This objective method of standardizing the image orientation should be useful for anatomical localization and clinical interpretation of these images.