Validation of an image quality index: its correlation with quality control parameters

Objective and Rationale: To validate a new image quality index (the Performance Index: PI) that assesses detectability of simulated lesions with a phantom. This index, presumably must depend markedly on quality control (QC) parameters as tomographic uniformity (Unif), Centre of Rotation (COR) and Spatial resolution (FWHM). The simultaneous effects of the QC parameters may explain much of the variation in the PIs; i.e. they may be predictors of the PI values. Methods: An overall performance phantom containing 3 sections was used. The first uniform section was used to determine tomographic uniformity. From the analysis of the slices corresponding to the second section containing 8 cold cylindrical simulated lesions of different diameters (range 7 mm - 17 mm), the number of true and false positives are determined and from these a new Performance Index (PI) is defined as the ratio between the positive predictive value and the sensitivity (expressed as its complement adding a constant to avoid a singularity). A point source located on the top of the phantom was used to determine the Centre of Rotation and the Spatial Resolution expressed by the FWHM in mm. 40 nuclear medicine labs participate at the survey. Standard multiple regression analysis between the Performance Index, as dependent variable, and FWHM, COR and Unif as independent variables was performed to evaluate the influence of the QC parameters on the PI values. Results: It is shown that resolution and COR are both predictors of the PIs, with statistical significance for the multiple correlation co-efficient R. However the addition of the variable tomographic uniformity to the model, does not improve the prediction of PIs. Moreover, the regression model lacks overall statistical significance. Regression summary for dependent variable Performance Index is presented. Conclusions: We confirm that the new Performance Index (PI), depends on QC parameters as COR and Spatial resolution. Those labs whose PIs are out of line with the prediction based on these QC parameters, may suggest other factors of importance affecting detectability, besides them