Generating harmonized SUV within the EANM EARL accreditation program: software approach versus EARL-compliant reconstruction.

Background: Evolutions in hardware and software PET technology, such as point spread function (PSF) reconstruction, have been shown to improve diagnostic performance, but can also lead to important device-dependent and reconstruction-dependent variations in standardized uptake values (SUVs). This may preclude the multicentre use of SUVs as a prognostic or diagnostic tool or as a biomarker of the early response to antineoplastic treatments. This study compared two SUV harmonization strategies using a newer reconstruction algorithm that improves lesion detection while maintaining comparability with older systems: (1) the use of a second reconstruction compliant with harmonization standards and (2) the use of a proprietary software tool (EQ.PET).

Methods: PET data from 50 consecutive non-small cell lung cancer patients were reconstructed with PSF reconstruction for optimal tumor detection and an ordered subset expectation maximization (OSEM3D) reconstruction to mimic a former generation PET. An additional PSF reconstruction was performed with a 7 mm Gaussian filter (PSF, first method), and, post-reconstruction, the EQ filter (same Gaussian filter) was applied to the PSF data (PSF, second method) for harmonization purposes. The 7 mm kernel filter was chosen to comply with the European Association of Nuclear Medicine (EANM) standards. SUVs for all reconstructions were compared with regression analyses and/or Bland-Altman plots.

Results: Overall, 171 lesions were analyzed: 55 lung lesions (32.2%), 87 lymph nodes (50.9%), and 29 metastases (16.9%). In these lesions, the mean PSF/OSEM3D ratios for SUV and SUV were 1.02 (95% CI: 0.93-1.11) and 1.04 (95% CI: 0.95-1.14), respectively. The mean PSF/OSEM3D ratios for SUV and SUV were 1.01 (95% CI: 0.91-1.11) and 1.04 (95% CI: 0.94-1.14), respectively. When comparing PSF and PSF, Bland-Altman analysis showed that the mean PSF/PSF ratios for SUV and SUV were 1.01 (95% CI: 0.96-1.06) and 1.01 (95% CI: 0.97-1.04), respectively.

Conclusion: The issue of reconstruction dependency in SUV values that hampers the comparison of data between different PET systems can be overcome using two reconstructions for harmonized quantification and optimal diagnosis or using the EQ.PET technology. Both technologies produce similar results, EQ.PET sparing reconstruction and interpretation time. Other manufacturers are encouraged to either emulate this solution or to produce a vendor-neutral approach.