The experiments were carried out on three camera heads of different crystal thicknesses (9.5, 15.9, and 25.4 mm) with large rectangular field of views.
Reference measurements of the multiple window spatial registration error were made with a 67Ga source (activity at the time of measurement ca. 18 MBq). The vial containing the 67Ga was placed in the standard lead collimator as defined by the NEMA standard (Fig. ) and placed at five arbitrary positions (centre, +x, -x, +y and -y) on the detector surface, with the ± x and ± y positions at about 2/3 of the distance between the centre and the respective border of the field of view. Only 5 positions were measured, compared to the NEMA recommended 9 positions. To ensure accurate and reproducible positioning, a template made of Perspex with machined recesses into which the collimator accurately fitted was used. The 133Ba point source is a standard commercial product (Spectrum Techniques, Oak Ridge, TN). The point source with a circular radioactive area of less than 0.2 mm diameter is embedded in the centre of a round plastic disk with a diameter of 26 mm and a total thickness of 2.7 mm. It had an activity of about 300 kBq at the time of measurements. The point source was imaged at exactly the same five positions using the same template with additional concentric recesses to fit the collimators for accurate positioning.
Figure 1 Cylindrical lead collimator for multiple window spatial registration measurement according to  showing liquid 67Ga source inside.
The collimators for the 133Ba point source were made of lead with dimensions given in Fig. . Collimators with central bores of 2, 3, 4 and 5 mm were produced. On the upper side of the collimators was a 0.5 mm thick recess tightly fitting the point source disk. The geometrical positioning of the source was thus ensured to be accurate within 0.1 mm. The weight of a collimator was 102 g.
Mini collimator made of lead showing 133Ba point source in plastic disc on top. Collimator hole sizes were 2, 3, 4 and 5 mm.
For each position and collimator geometry two images for the low energy photons and the high energy photons of either the 67Ga or 133Ba source were acquired for 120 s in a 512 × 512 matrix, pixel size 1.1 mm. The peak energy settings were 93 and 300 keV for 67Ga and 80 and 356 keV for 133Ba, respectively, with all energy windows set to 20%. The position of the source in an image was determined according to the NEMA algorithm by calculating the centre of gravity of the intensity within a square region with a side length of four times the full width at half maximum (FWHM) of the peak. To avoid potential errors from differences in region positioning a program written in Matlab (Version 2006a, The Mathworks Inc., Natick, MA) was developed in which the approximate position of the point source was first identified manually, followed by a fully automatic placement of the region of interest and calculations of the registration offsets. In order to identify optimum collimator dimensions, initial experiments were made using collimator bore diameters of 2, 3, 4 and 5 mm as well as the 133Ba source without collimation. Since the collimators with 2 and 3 mm bores were inferior to the collimators with larger holes with respect to sensitivity (see results section), the localization comparisons with the 67Ga values were then made with the collimators with a hole size of 4 and 5 mm and with the 133Ba source without collimation.
In order to facilitate the comparison of the results, the differences between low and high energy positions for the two radionuclides were displayed visually. The similarity between the locations of the two radionuclides were analyzed further by performing regression analyses using the statistical software SPSS (Version 14.0.1, SPSS Inc., Chicago, IL). Statistical significance was assumed for a p value < 0.05. A Bland-Altman analysis [7
] was performed for statistically significant regressions.
Reproducibility was tested both for the 133Ba method using the 5 mm collimator and the 67Ga method by acquiring low and high energy images in one template position. For each configuration 10 measurements were carried out. Between each measurement the source with collimator was completely removed and positioned anew. Reproducibility and accuracy of the centre of the 133Ba source were tested by acquiring low and high energy images in one template position, for the 5 mm collimator. Eight acquisitions were performed. Between each of these acquisitions the point source was rotated by 45°.