Image quality and quantitation accuracy of penalized-likelihood reconstruction method for ¹⁸F and ⁹⁰Y

One application of PET/CT is diagnosis of tumours using ¹⁸F-FDG as a radiotracer. Early detection of small lesions is imperative to improve diagnosis. Recently, PET/CT was proposed to image ⁹⁰Y, (which has a small number (32ppm) of positrons emitted from internal pair production), as a tool for verification of radiotracer distribution and dosimetry for radioembolization therapy. Ordered Subset Expectation Maximization (OSEM), the standard image reconstruction algorithm in the clinic, creates mottled images making small lesion detection for ¹⁸F and ⁹⁰Y PET imaging difficult. The penalized-likelihood (PL) reconstruction technique with a relative difference penalty (RDP) based on a block sequential regularized expectation maximization (BSREM) numerical optimizer can create smooth regions without compromising the edges. The objective of this thesis was to quantitatively compare small-lesion detectability and quantitation accuracy of BSREM with the OSEM algorithm. This was performed with and without time-of-flight (TOF) information in both phantom and patient studies for ¹⁸F and ⁹⁰Y. For ¹⁸F phantom experiments, standard International Electrotechnical Commission (IEC) spheres were scanned using 4 different signal-to-background (SBR) ratios (=Air, Water, 8 and 4). Four smaller spheres were scanned to evaluate the performance of the algorithms in detecting small objects. Image quality was analyzed using 60 lung and head/neck tumour patient scans. Images reconstructed using TOF-BSREM showed an increase of signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) for small spheres by 16±3% and 35±2% compared to OSEM. For the patient scans, TOF-BSREM did not show any improvements of CNR but an increase of 14±4% for SNR. The recovery of ¹⁸F activity in large spheres, was underestimated by 10%, while for small spheres, accuracy was even more decreased due to partial volume effect. For ⁹⁰Y scans, the same spheres were scanned with a SBR of 7.5. ⁹⁰Y images reconstructed using TOF-BSREM showed an improvement of SNR and CNR by 89±7% and 96±8% respectively. TOF-BSREM improved quantitation accuracy to 78% in contrast to 65% with OSEM, when physical boundaries of sphere were taken as the region-of-interest. In conclusion, SNR, CNR and quantitation accuracy improved for small spheres when using the TOF-BSREM algorithm.Science, Faculty ofPhysics and Astronomy, Department ofGraduat