Evaluation of a Monte Carlo scatter correction in clinical 3D PET

Phantom and patient data were used to compare performance of a one-iteration Monte Carlo scatter correction (MC-SC-1i) for 3D PET, a vendor-supplied one-iteration single scatter model-based correction (SSS-1i) for 3D PET, unscatter-corrected 3D PET (No-SC), a SSS-1i followed by Monte Carlo scatter correction as a second iteration (MC-SSS) for 3D PET, and a convolution-subtraction scatter correction for 2D PET in terms of quantitative accuracy and lesion detectability. ROI analysis showed 2D PET images were more accurate than 3D, particularly for large phantoms, and MC-SSS corrected 3D PET images were more accurate than SSS-1i corrected 3D PET images for this data set. 2D and 3D PET images were reconstructed from 59 patient data sets. Bias of 3D PET images with respect to 2D images was determined using Corresponding Intensity Variance. 3D PET uncorrected images overestimated activity by ∼50% (smallest patients) to ∼150% (largest patients). The average absolute bias of SSS-1i corrected images (16%) was twice that of MC-SSS (8%) and more dependent on patient size. Lesion detection sensitivity in these patient images was evaluated using a Channelized Hotelling Observer. Scatter corrected 3D PET images performed ∼10% better than uncorrected 3D PET images for smaller patients. Slightly better lesion sensitivity was seen for large patients in images reconstructed using SSS-1i (CHO-SNR=2.23±0.29) compared to MC-SSS (2.08±0.27) and uncorrected images (2.02±0.23).