Comparison of phantoms for PET scanner harmonization

Objectives: PET scanner harmonization techniques have been developed to reduce quantitative discrepancies arising from hardware and software differences between scanners. Nearly all currently proposed harmonization strategies require the NEMA IEC Body Phantom to be scanned on all scanners of interest. We investigated if more simple and commonly-used phantoms could instead be used in scanner harmonization.

Methods: Three different PET phantoms were evaluated: the NEMA IEC Body Phantom, the ACR-approved image quality phantom, and a uniformity phantom. The phantoms were scanned under identical clinically-relevant conditions on 2 different PET/CT scanners at our institution: the GE Discovery VCT and GE Discovery 710 scanner. The Discovery VCT was selected as the reference scanner due to its inferior image quality. For each phantom scan, numerous reconstructions were processed using different post-filter widths and reconstruction settings. Dissimilarity scores, D(s), were calculated for each reconstructed image, where D(s) was defined as the squared difference in SUVmax between the two scanners summed over all lesion sizes. The reconstruction settings with the lowest D(s) were selected as the harmonized reconstruction. Due to the absence of lesions in the uniformity phantom, noise-based harmonization was used instead, where D(s) was defined as the squared difference in noise values, with noise defined as the standard deviation of voxel values within a large ROI. Harmonization techniques were compared in an example lung cancer patient who received 5 PET/CT scans on the two scanners over the course of 13 months.

Results: Harmonization reduced differences in SUVmax between scanners from an average of 20 ± 11% to within 5-7%, regardless of the phantom. The D(s) scores were highly correlated between ACR phantom harmonization and NEMA phantom harmonization (R=0.99), indicating that scanner harmonization using either phantom produced nearly identical results. D(s) scores for the uniformity phantom (ie, noise-based harmonization) were not strongly correlated with those from the ACR phantom (R=0.53) nor the NEMA phantom (R=0.47), yet the single top-performing harmonized reconstruction nonetheless reduced differences in SUVmax between the scanners to within 5-7%. Application of the different harmonization methods to the example patient also resulted in minor differences in SUVmax between the harmonization methods (3-4%), whereas differences between the harmonized and original non-harmonized images were large (60-80%).

Conclusion: Scanner harmonization for SUVmax can be performed using either the NEMA phantom or the ACR phantom. Furthermore, we found that noise-based harmonization using a simple uniformity phantom had comparable results to the other phantoms in our 2 GE PET/CT scanners, reducing inter-scanner differences in SUVmax to within 5-7%. A comparison of the different harmonization phantoms in a greater variety of PET scanner models is warranted. Research Support: This project was supported by the Departments of Radiology and Medical Physics, University of Wisconsin.