The impact of reconstruction and scanner characterisation on the diagnostic capability of a normal database for [123I]FP-CIT SPECT imaging

BACKGROUND The use of a normal database for [123I]FP-CIT SPECT imaging has been found to be helpful for cases which are difficult to interpret by visual assessment alone, and to improve reproducibility in scan interpretation. The aim of this study was to assess whether the use of different tomographic reconstructions affects the performance of a normal [123I]FP-CIT SPECT database and also whether systems benefit from a system characterisation before a database is used. Seventy-seven [123I]FP-CIT SPECT studies from two sites and with 3-year clinical follow-up were assessed quantitatively for scan normality using the ENC-DAT normal database obtained in well-documented healthy subjects. Patient and normal data were reconstructed with iterative reconstruction with correction for attenuation, scatter and septal penetration (ACSC), the same reconstruction without corrections (IRNC), and filtered back-projection (FBP) with data quantified using small volume-of-interest (VOI) (BRASS) and large VOI (Southampton) analysis methods. Test performance was assessed with and without system characterisation, using receiver operating characteristics (ROC) analysis for age-independent data and using sensitivity/specificity analysis with age-matched normal values. The clinical diagnosis at follow-up was used as the standard of truth. RESULTS There were no significant differences in the age-independent quantitative assessment of scan normality across reconstructions, system characterisation and quantitative methods (ROC AUC 0.866-0.924). With BRASS quantification, there were no significant differences between the values of sensitivity (67.4-83.7%) or specificity (79.4-91.2%) across all reconstruction and calibration strategies. However, the Southampton method showed significant differences in sensitivity between ACSC (90.7%) vs IRNC (76.7%) and FBP (67.4%) reconstructions with calibration. Sensitivity using ACSC reconstruction with this method was also significantly better with calibration than without calibration (65.1%). Specificity using the Southampton method was unchanged across reconstruction and calibration choices (82.4-88.2%). CONCLUSIONS The ability of a normal [123I]FP-CIT SPECT database to assess clinical scan normality is equivalent across all reconstruction, system characterisation, and quantification strategies using BRASS quantification. However, when using the Southampton quantification method, performance is sensitive to the reconstruction and calibration strategy used.