Multicenter Comparison of Calibration and Cross Calibration of PET Scanners

MATERIALS AND METHODS

From the institutions participating in the clinical multicenter study, 23 scanners of 7 types were investigated: 15 ECAT EXACT, 3 ECAT EXACT HR+, 1 ECAT 951, and 1 ECAT ART (CTI, Knoxville, TN/Siemens Medical Systems, Inc., Hoffman Estates, IL); 1 Advance (General Electric Medical Systems, Milwaukee, WI); and 1 QUEST and 1 PENN PET (UGM Medical Systems, Inc., Philadelphia, PA). Of these, 16 were operable in both 2-dimensional (2D) and 3-dimensional (3D) modes, 4 were operable in 2D mode only, and 3 were operable in 3D mode only. This mix fairly represents the installed base of PET in Germany.

In a first step, the accuracy of the dose calibrators at all sites was examined by measuring ⁶⁸Ge standards (AEA Technology QSA GmbH, Braunschweig, Germany) with 42 MBq, 4.03 MBq, and 0.383 MBq, respectively, in 2 mL, calibrated to an overall uncertainty of ±5%. The prechecked dose calibrator was used to homogeneously fill a cylindric phantom of well-known volume (20 cm in diameter and 20 cm in length) with an aqueous solution of ¹⁸F (0.025 MBq/mL for scanners of 2D mode and 0.006 MBq/mL for scanners of 3D mode, reflecting associated differences in sensitivity). From each solution, 2 aliquots (2 mL) were withdrawn and counted in the well counter. Subsequently, the phantom was scanned for 60 min using either mode of data acquisition. The complete procedure, including filling of the phantom, was repeated twice.

After application of all corrections (e.g., scatter, dead time, random coincidences, attenuation, and detector normalization) as provided by the standard software of each scanner, images were reconstructed by applying filtered backprojection using a ramp filter with Nyquist frequency cutoff. If available, 3D data were reconstructed using a linear reprojection algorithm with subsequent 3D filtered backprojection (4). In all cases, attenuation correction was performed using measured cold transmission data of high statistical precision (scan duration, 30 to 60 min, depending on actual source strength).

All data were analyzed by the same person using a standardized procedure. After visual inspection for image artifacts, plane-to-plane sensitivity variations, and nonuniformities, the calibration of the scanner and its cross calibration to the dose calibrator were checked by comparing the image activity concentration in regions of interest (15 cm in diameter and centered on the phantom) with phantom activity concentration, as calculated from the activity measured by the dose calibrator and phantom volume. Finally, the cross calibration of the well counter to the scanner was tested by comparing well counter measurements of the aliquots with the corresponding image activity concentration.

Our aim was not to compare and stage different types of scanners or manufacturers—although there are differences—but to ensure the quantitative accuracy of all participating instruments. Therefore, the results from the different PET scanners and their associated periphery were documented graphically in a random sequence established beforehand.

RESULTS

Scanner Calibration and Cross Calibration to Dose Calibrator

In all scanners, differences in image activity concentration between both repeated phantom measurements were <3% in both 2D and 3D acquisition modes. For this reason, the results of just 1 representative phantom measurement are shown (Fig. 1). Errors from PET concentration measurements, as compared with dose calibrator data, were <5% for 11 of 20 scanners operated in 2D mode and <5% for 9 of 19 scanners operated in 3D mode. For 5 scanners in 2D mode and 4 scanners in 3D mode, the error was approximately 10%. For these, a new cross calibration was recommended to improve accuracy.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1.

Percentage error of activity concentrations derived from PET images (white columns) for 2D (A) and 3D (B) acquisitions. Reference value is activity concentration as determined from dose calibrator measurement of activity injected and phantom volume. Black columns represent percentage errors after maintenance or corrective procedures. Scanners 6, 8, 9, and 23 could be operated in 2D mode only, and scanners 15, 20, and 21 could be operated in 3D mode only. Values of scanner 21 could not be evaluated. Initial values of scanner 14 for 2D and 3D and scanner 4 for 3D were not available.

Because of incorrect calibration, scanners 2, 3, 4, 11, and 14 did not initially qualify for quantitative data analysis. After the problems had been identified and fixed, the measurements were repeated. The success of these corrections is also shown in Figure 1 (black bars). The identified problems ranged from those that were technical, such as an inoperative dead-time correction because of lost normalization software components (scanner 3) and a failed decay correction because of faulty internal time-zone settings (scanner 2), to those that related to user handling errors, such as a failure to perform calibration at all (scanner 14), an incomplete ¹⁸F cross calibration (scanner 11), and a phantom-software mismatch (scanner 4). Finally, scanner 21 had to be excluded from quantitative analysis because of major hardware and software problems that could not be corrected during the time frame of the study.

Cross Calibration to Well Counter

The errors in activity concentrations derived from PET images with respect to well counter measurements of the aliquots are illustrated in Figure 2. This is the direct link between blood and tissue concentration needed for physiologic modeling. Even if the same well counter is used for both 2D and 3D measurements, errors in scanner calibration will influence the results.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 2.

Mean percentage error of activity concentrations derived from PET images with respect to values measured in well counters for 2D (hatched columns) and 3D (dotted columns) modes. Respective data for well counters 2, 9, and 21 are missing, because well counters were not available at these centers. Error of well counter 8 is near zero and cannot be seen on this scale.

Most of the tested devices deviated by <10%; nearly half were even better, deviating by <5%. Pronounced errors in accuracy, indicating possible malfunction of the instrument, were found in only a few devices (well counters 5, 12, 13, 16, 18, and 20). This group of instruments could be subdivided into 2 classes, those with errors in accuracy and those with, in addition, large variations between individual measurements (data not shown).

DISCUSSION

In quantitative PET, the image count rate per volume (representing tissue activity concentration) is related to the injected dose (e.g., determination of standardized uptake values) and to the blood activity concentration (physiologic modeling). Therefore, careful cross calibration between the PET scanner, dose calibrator, and well counter is essential.

The calibration accuracy of the scanners in this study was examined by comparing phantom data between institutions. When the accuracy was found to be degraded, we tried to identify the source of the error. We do not describe a new method of calibrating PET scanners, but we investigated the accuracy of the calibration process as performed in the field and in the context of a multicenter study. To facilitate analysis of errors and pooling of data from different institutions, we also checked the absolute accuracy of the dose calibrators using a set of certified standards. Nevertheless, the tests that could be performed in the framework of this study cover only the basic prerequisites for quantifying PET activity concentrations in vivo (5).

All investigated PET scanners were suitable for visual analysis of data, as judged by daily quality control and visual inspection of sinograms and reconstructed images. However, some scanners could not be used for quantitative studies without prior corrective maintenance. Quality control, although a prerequisite for operation of PET scanners, does not guarantee the accuracy of scanner calibration.

The calibration chain starts with the dose calibrator, whose accuracy is therefore of fundamental importance. In this respect, the test outcome was quite satisfying. Nearly all instruments had a <10% deviation, and most were even better (<5%). The outliers were either defective because of age or contaminated—problems mandatory to be solved. Less satisfactory were the tests on scanner calibration and cross calibration to the dose calibrator. At the first attempt, only 50% of the devices showed <5% deviation, but after the problems had been identified and solved, the accuracy of most of the remaining devices could be improved to ≤10%. Only 1 scanner could not be fixed at all, because of unresolvable problems with hardware and software.

The sources of error ranged widely—from more sophisticated problems, such as misalignment of processor clocks and loss of some components of the normalization and correction software data, to simple user handling errors (these being the majority), such as failure to perform calibration at all, faulty specification of the phantom activity concentration, and errors in performing the calibration and cross calibration.

Only a few of the centers actually used measurement of blood samples as an input function for physiologic modeling and, therefore, needed a cross-calibrated well counter. Nevertheless, when this type of device was available it was checked. So the relatively large errors for some of the tested instruments may be explained by infrequent use and, therefore, no need for cross calibration.

CONCLUSION

For clinical multicenter studies relying on quantitative analysis of PET data, the calibration of scanners must carefully be checked beforehand. By thorough application of straightforward, standard procedures, an accuracy of at least 5%–10% could be achieved for nearly all the dedicated PET scanners tested.

However, the measurements and readjustments performed during this study reflected the status of the devices at that time. Permanent monitoring by the participating institutions is needed to maintain this condition.