Abstract
2313
Objectives Harmonized quantitation of PET measured lesions would enhance the statistical power of many oncology multi-center clinical trials. This work assesses the feasibility of identifying optimized PET/CT scanner model-specific reconstruction parameters to generate identical recovery coefficient (RC) curves with minimal bias and variability. It also investigates volume of interest (VOI) strategies to minimize noise impact on RCs while maintaining accuracy and minimizing bias.
Methods A NEMA NU2 Image Quality phantom with standard sphere set was imaged on a Siemens Biograph 40 and an older generation Biograph Duo for 30 minutes to achieve robust statistics for reconstructions. Images were reconstructed (2D OSEM, 256x256) using an array of iterations, subsets, and smoothing filters. RC curves were generated using max pixel values, 2 pixel cube, and 3 and 5 pixel diameter spheres. Promising reconstruction parameter sets from each scanner that generated similar RC curves were identified. The phantom was reimaged for a 6x5min. dynamic sequence on both scanners and reconstructed using the identified harmonized reconstruction parameter sets to assess variance of RC values. Priority was given to raising quantitative performance of the older scanner with increased iterations/subsets and decreasing filter FWHM.
Results Scanner specific reconstruction parameters were identified for the Biograph 40 and Duo that generated optimized harmonization between RC curves for each of the VOi strategies (table below).
Conclusions Optimized and harmonized reconstructions for two different vintage scanners were achieved by driving higher resolution performance through increasing iterations/subsets and decreasing Gaussian filter FWHM on the earlier generation scanner. A larger FWHM filter on the newer PET/CT proved necessary for harmonization. Using larger VOI's as the RC curve metric helped dampen the effects of the increased noise from the higher iterations, minimizing bias