Iterative reconstruction and its impact on tumor quantitation of Ga-68 PET/CT

Objectives Ordered Subset Expectation Maximization (OSEM) is widely used in research and clinical PET/CT applications. These methods approximate the true distribution of the events, by maximizing a posteriori likelihood function through an iterative process including segmentation into subsets. To determine the impact of iterations and subsets on quantitative PET, we evaluated NEMA phantom images and clinical research patients with neuroendocrine tumors.

Methods A NEMA 2007/IEC 2008 phantom (spheres: 10-37mm diameters) filled with Ga-68 (1.12uCi/ml per sphere and 0.191uCi/ml background) was imaged on a Siemens Biograph mCT for 6 min. Reconstructions were performed using Filtered Backprojection (FBP) and OSEM, using a range of iterations (1, 2, 3, 6, 9, and 12) and subsets (4, 8, 12, and 24). All images were filtered (2mm FWHM Gaussian) and attenuation corrected. Similarly, patients with somatostatin positive tumors were administered 2.5±0.2 MBq/kg [Ga -68]-DOTA-NOC (n=6) and whole body PET/CT images were acquired and reconstructed as described above. In all cases, FBP images were segmented via region growing and were quantified for total SUV across all reconstruction methods. For patients, tumors in 3 size categories (2.4±0.4, 11.8±2.0, and 107±41.3 ml) were analyzed as described above.

Results Analysis of phantom and patient data showed a statistically significant non-linear dependence with both numbers of iterations and subsets across all volumes studied when comparing OSEM to FBP. The relationship between iterations, or subsets, and relative error rates decrease asymptotically with increasing parameter values. In all cases, relative error were object size dependent and ranged from -53.6% to 9.6%, and from -60.1% to 6.7%, for phantom and patient data, respectively.

Conclusions Combined these data indicate that OSEM can adversely affect image quantitation. Moreover, given the size dependency of errors, the detection of small tumors may be impacted resulting in underestimates of true tumor burden.