The impact of image reconstruction bias on PET/CT 90Y dosimetry after radioembolization.

UNLABELLED PET/CT imaging after radioembolization is a viable method for determining the posttreatment (90)Y distribution in the liver. Low true-to-random coincidence ratios in (90)Y PET studies limit the quantitative accuracy of these studies when reconstruction algorithms optimized for traditional PET imaging are used. This study examined these quantitative limitations and assessed the feasibility of generating radiation dosimetry maps in liver regions with high and low (90)Y concentrations. METHODS (90)Y PET images were collected on a PET/CT scanner and iteratively reconstructed with the vendor-supplied reconstruction algorithm. PET studies on a Jaszczak cylindric phantom were performed to determine quantitative accuracy and minimum detectable concentration (MDC). (90)Y and (18)F point-source studies were used to investigate the possible increase in detected random coincidence events due to bremsstrahlung photons. Retrospective quantitative analyses were performed on (90)Y PET/CT images obtained after 65 right or left hepatic artery radioembolizations in 59 patients. Quantitative image errors were determined by comparing the measured image activity with the assayed (90)Y activity. PET images were converted to dose maps through convolution with voxel S values generated using MCNPX, a Monte Carlo N-particle transport code system for multiparticle and high-energy applications. Tumor and parenchyma doses and potential bias based on measurements found below the MDC were recorded. RESULTS Random coincidences were found to increase in (90)Y acquisitions, compared with (18)F acquisitions, at similar positron emission rates because of bremsstrahlung photons. Positive bias was observed in all images. Quantitative accuracy was achieved for phantom inserts above the MDC of 1 MBq/mL. The mean dose to viable tumors was 183.6 ± 156.5 Gy, with an average potential bias of 3.3 ± 6.4 Gy. The mean dose to the parenchyma was 97.1 ± 22.1 Gy, with an average potential bias of 8.9 ± 4.9 Gy. CONCLUSION The low signal-to-noise ratio caused by low positron emission rates and high bremsstrahlung photon production resulted in a positive bias on (90)Y PET images reconstructed with conventional iterative algorithms. However, quantitative accuracy was good at high activity concentrations, such as those found in tumor volumes, allowing for adequate tumor (90)Y PET/CT dosimetry after radioembolization.