Accuracy of partial volume correction: CT versus PET for tumor volume definition

Abstract

1725

Objectives: To compare accuracy of partial volume correction (PVC) when the tumor volume of interest (VOI) is defined by CT or PET. Methods: A mathematical anthropomorphic phantom was used containing spherical tumors of 10 to 40 mm diameter and various elliptical and heterogeneous tumors. Background activity concentration was set to 5 kBq/cc and tumor-to-background ratio to 5 or 10. Simulations included PET spatial resolution (~7 mm FWHM) and Poisson statistics (5 min, 3D scan). PET tumor VOI was based on a 50% isocontour of maximum standardized uptake value (SUV), with background correction [JNM 45:1519-27; 2004]. For CT tumor definition, the simulated VOI itself was used, i.e. it was assumed that CT provided exact tumor definition. Shifts of 0.0, 2.5 and 5.0 mm between CT and PET data were applied to simulate small displacements (i.e. patient movements/breathing) between CT and PET. Finally, SUV data were corrected for PVC using both CT- or PET-VOI, together with a scanner resolution model and a convolution based PVC method. Results: For homogeneous tumors larger than 20 mm diameter, use of CT- and PET-VOI provided accurate (PVC) corrected SUV (error < 10%). For smaller tumors (10 mm diameter), however, use of PET-VOI resulted in 60% underestimation of PVC SUV. Similar results were found for small (short axis < 20 mm) non-spherical tumors. In case of heterogeneous tumors, use of PET-VOI resulted in 50% under- to 100% overestimations of PVC SUV. In the latter case, use of CT-VOI was able to provide a PVC SUV accuracy within 20%. Displacement between CT and PET data of more than 2.5 mm caused large underestimations of PVC SUV of up to -50%, thereby completely offsetting PVC itself. Conclusions: Assuming that CT data can provide accurate tumor VOI, PVC based on this VOI outperforms PVC based on PET-VOI for small tumors (<20 mm) and in case of tumor heterogeneity. PVC based on CT-VOI is, however, very sensitive to small displacements between PET and CT, thereby dictating the need for rigid patient immobilization and use of respiratory gating for both PET and CT data.