Development and testing of a formalism to identify harmonized and optimized reconstructions for PET/CT in clinical trials.

Objectives Differing quantitative performance characteristics of PET/CT scanner models and reconstructions confound the ability to best combine and compare data. Here we develop and test a formal method for identifying quantitatively similar scanner and reconstruction specific performances through simultaneous contrast recovery coefficient (CRC) curve harmonization (CRCs aligned) and optimization (CRCs close to 1.0).

Methods A NEMA IQ phantom was imaged for 30 minutes at 9.7:1 contrast using 12 sphere sizes 8.5-44mm (6 at a time). Images were reconstructed using at least 6 different clinically relevant reconstruction parameter sets per scanner. CRC curves were generated for SUV_max, SUV_peak and SUV_mean. 5 PET/CT models (11 scanners) spanning major manufactures were evaluated. Summed absolute differences between all combinations of CRC curves were calculated to identify reconstruction parameter sets that were best harmonized. Summed differences for all combinations of CRC curves from the ideal CRC=1 were calculated to assess optimization. The smallest weighted combination of the two parameters determined the scanner-specific parameter sets that best aligned quantitative performance.

Results A 7mm Gaussian post-reconstruction filter applied to GE and Siemens systems best drove harmonization of SUV_max CRCs across Philips, GE and Siemens systems. Generally the same parameter sets that harmonized SUV_max also harmonized SUV_peak and SUV_mean. Applying additional optimization criteria favoring CRCs closer to 1.0 can be achieved, but at the expense of most favorable harmonization.

Conclusions A formal method to identify PET/CT model specific reconstruction parameter sets that both harmonize and optimize quantitative performance was developed and tested using data from the NEMA IQ phantom with 5 common PET/CT platforms. The method is scalable and can be applied to the full set of available PET/CT scanner models.

Research Support Research reported in this abstract was supported by the National Cancer Institute of the National Institutes of Health under Award Number R01CA169072