Evaluation of algorithms for the registration of 99Tcm-HMPAO brain SPET studies.

ABSTRACT

Five functions of merit were used for the registration of 99Tcm-HMPAO brain SPET studies. The correlation coefficient (COR), the stochastic sign change (SSC), the standard deviation of ratios (SDR), the sum of the absolute differences (SAD) and a new function based on a local correlation coefficient (LOC) were tested in the registration of photic neuroactivation (ACT), epilepsy (EPL) and Wada (WAD) SPET studies. The comparison included simulated and real studies. The translation error in registration was 0.1 +/- 0.1 pixels (mean +/- S.D.) for all functions of merit for the complete set of simulated studies (10 runs for each ACT, EPL and WAD). For rotation, LOC yielded the best results with a mean error of 0.3 +/- 0.2 degree and a maximum error of 0.6 degree. Slightly higher errors were found with SAD (0.4 +/- 0.2 degree, maximum 1.0 degree) and COR (0.5 +/- 1.0 degree, maximum 1.0 degree). The highest errors were found with SDR (0.8 +/- 1.0 degree, maximum 4.8 degrees) and SSC (0.8 +/- 1.1 degrees, maximum 4.7 degrees). The results obtained from five real studies of ACT, of EPL and of WAD were in agreement with the findings from the simulated studies, thus confirming the robustness of LOC, SAD and COR for the registration of 99Tcm-HMPAO brain SPET studies.