Nuclear Medicine Communications 11, 95-101 (1990) Reproducibility and action levels for gamma camera uniformity K.C. YOUNG, K. KOURIS, M. AWDEH and H.M. ABDEL-DAYEM Department of Nuclear Medicine, Kucoait University, P.O. Box 24923, Safat, Kuwait 13110 Received in revised form 23 October 1989 Summary Deciding on the action level for gamma camera non-uniformity is difficult because the reproducibilities of quality control measurements and service adjustments are usually unknown, This work evaluated the reproducibilities of integral uniformity (IU), differential uniformity (DU) and the corrected relative standard deviation (CRSD). The latter was calculated by removing from the relative standard deviation of the pixel counts the component due to statistical fluctuations. The reproducibility of each parameter was evaluated by analysing 10 intrinsic flood acquisitions with total counts of 2, 5, 10 and 30 million. All three parameters were less reproducible at the lower count densities, but as expected IU and DU also showed higher mean values. CRSD was consistent and highly reproducible, at all count densities. At 10 million counts CRSD had a coefficient of variation (COV) of 1.3% which was a five-fold improvement over the 6.6% and 6.1% found for IU and DU, respectively. The relative sensitivity of 1U, DU and CRSD was compared in monthly measurements on 10 gamma cameras over one year. No significant difference in relative sensitivity was demonstrated: a change in camera performance produced about the same percentage change in each parameter. The precision with which service engineers adjust gamma cameras was also assessed by measuring the uniformity of 10 gamma cameras immediately after service adjustment at monthly intervals over one year. Finally, general action levels were defined for IU, DU and CRSD at 7%, 5% and 2.5% respectively. Introduction Quantitative measurements of gamma camera uniformity are used as part of routine quality control on camera performance, acceptance testing and evaluation of improvements in camera design. A wide variety of methods of quantifying uniformity have been described in the literature [1-10], The relative merits of the different methods are not well established. The four criteria required of a performance parameter have been described [1] as relevance, simplicity, sensitivity and repro- ducibility. It is the objective of this paper to assess the reproducibility and relative 0143-3636/90 $03.00+.12 © 1990 Chapman and Hall Ltd.96 YOUNG, KOURIS, AWDEH and ABDEL-DAYEM sensitivity of three uniformity parameters and to set action levels for a quality control programme. Since the NEMA protocols were introduced [4] their methods for measuring integral uniformity (IU) and differential uniformity (DU) have been adopted widely and are implemented as quality control parameters at many centres, although there are sometimes practical difficulties in reproducing their methods exactly. IU and DU were assessed in this study using a procedure similar to NEMA’s. The third parameter evaluated was the corrected relative standard deviation (CRSD). It is calculated by removing from the relative standard deviation of the pixel counts the contribution due to statistical fluctuations. This parameter is identical to the uniformity index recommended by Cox and Diffey 9] and the AAPM [10], but unfortunately it does not appear to have gained widespread use. One of the reasons for wanting to know the reproducibility of uniformity parameters is as an aid to defining action levels for quality control measurements. Such an action level needs to be sufficiently above the mean that it will only be reached when the camera performance truly deteriorates and not by chance fluctuations in the uniformity measurement. Also an action level must be sufficiently above the mean that an engineer will be able to make an improvement. A knowledge of all these factors is necessary to define suitable action levels. Methods Uniformity analysis To assess uniformity, an intrinsic flood image was acquired in a 64X64 matrix using a °Tc™ point source at a distance greater than five times the useful field of view (UFOV) from the centre of the uncollimated crystal. A circular region of interest was inscribed just within the flood image to define the UFOV. Only pixels within this circle were used to assess the uniformity, TU was calculated, after a nine point spatial smooth as TU = {(max—min)/(max-+min)}<100% where max and min denote the maximum and minimum pixel counts within the UFOV. To calculate the DU a 6X6 pixel grid was moved through the UFOV of the smoothed image and the above formula was applied for each location of the grid. The maximum value obtained was the DU. (Note that this method is somewhat different from NEMA’s, where DU is the maximum difference between pixels in strips of 5 in the x and y directions separately.) The relative standard deviation RSD was found by calculating the standard deviation of pixel counts in the UFOV of the unsmoothed image and representing this fraction as a percentage of the mean pixel counts. The RSD includes a major contribution from the statistical fluctuations in the pixel counts. This contribution was calculated and eliminated, thereby obtaining a corrected RSD (CRSD) arising from the camera non-uniformity alone. This was done using the rules for combining errors, to derive the following equation: CRSD = V{RSD=(SREDY where the statistical relative standard deviation (SRSD) = 100/Vmean count.Reproducibility and action levels for gamma camera uniformity 97 Reproducibility and sensitivity In order to assess the reproducibility of the above three non-uniformity parameters, 10 sequential intrinsic floods were acquired on a gamma camera using a pre-set count of 2 million. The uniformity of each image was analysed and the mean and standard deviation calculated for each one of the quantitative parameters. This was repeated using pre-set counts of 5, 10 and 30 million. To assess the relative sensitivity of the parameters, scatter plots were made of DU and CRSD against IU. The data points used were those obtained for 10 gamma cameras following regular monthly preventive maintenance over a one year period. In each case, the slope of a linear fit to the data was determined using regression analysis. Variability in adjusted cameras In order to assess the size of a uniformity deterioration that an engineer could usefully adjust, the uniformity of cameras were assessed after their monthly retuning over a one year period. Action levels for each camera were defined as the mean adjusted value plus two standard deviations (6.0.) for each parameter, Results The results of the reproducibility measurements are shown in Table 1 and Fig. 1. Increasing the total counts improved the reproducibility of all the parameters and reduced the mean value of IU and DU. The mean value of CRSD remained constant. The coefficients of variation (COV = s.0.x100/mean) of IU, DU and CRSD at different count densities are shown in Table 2. For all three parameters the COV improved with count density as reproducibility was improved. However, CRSD had much better COV values than IU or DU. 0 s 8 2 5 6 u é 2 4 bu & = pe org 5 2 °S 050 2060 3000 4000 6000 ed00 7600 6000 8C00 10800 Td00 Mean Counts per Pixel Fig. 1, The means (+/- 8.0.) of 10 measurements of the uniformity parameters IU, DU and CRSD are plotted against the mean pixel count densities for intrinsic flood images with total counts of 2, 5, 10 and 30 million,