Quantitative comparison of automatic and interactive methods for MRI-SPECTimage registration of the brain based on 3-dimensional calculation of error

A wide range of techniques for registration of medical images has been devi sed in recent years. The aim of this study is to quantify the overall spati al registration error of 3 different methods for image registration: intera ctive matching, surface matching, and uniformity index matching as describe d by Woods. Methods: MRI and ethylcysteinate dimer-SPECT images of the brai n were registered for 15 patients. The matching error was assessed by deter mining intra- and interobserver variability of registrations. Quantificatio n of the registration error was based on the mean spatial distance of 5000 voxels between 2 image positions. The mean position after repeated registra tions in each patient was used as the gold standard. To evaluate the cohere nce of the 3 different registration methods, intermethod variability was de termined. Results: Interactive matching showed an intraobserver/interobserv er variability of 1.5 +/- 0.3 mm/1.6 +/- 0.3 mm (mean +/- SD). The time dem and for this method was 11 +/- 5 min. Surface matching revealed a variabili ty of 2.6 +/- 1.1 mm/3.8 +/- 1.0 mm and a time demand of 26 +/- 12 min. Rep roducibility of Woods' algorithm was 2.2 +/- 0.8 mm with a time demand of 9 +/- 3 min. In 4 of the 15 cases, Woods' method failed. The mean deviation between all 3 methods was 2.3 +/- 0.8 mm, Conclusion: With a suitable user interface, interactive matching had the lowest registration error. The infl uence of subjectivity was shown to be negligible. Therefore, interactive ma tching is our preferred technique for image fusion of the brain.