PET quantification inaccuracy of non-uniform tracer distributions for radiation therapy

The quantitative accuracy of PET has been measured for phantoms containing uniform activity spheres and cylinders, but has never been studied for non-uniformly varying tracer concentration (TC) distributions. For the correct delineation of tumors for radiation therapy (RT) using PET, the combined effect of all PET image degrading factors for realistic cases needs to be known. We investigate PET inaccuracy for a phantom with TC and attenuation varying non-uniformly in one direction. This ID model enables determination of the positional, angular and statistical effects on the quantification accuracy. We simulated a cylindrical uniform TC phantom containing a non-uniform TC insert using the GATE (Geant4 Application for Tomographic Emission) Monte Carlo code. The insert is made of thin parallel slabs with different TC, atomic composition and density. The phantom is simulated inside a validated model of the GE Discovery LS PET/CT scanner. The Software for Tomographic Image Reconstruction (STIR) was used to reconstruct the PET images, which were then analyzed by comparing the reconstructed to true TC recovery coefficients. Simulated images with and without attenuation (AC) and scatter and random events included and rejected were analyzed. The results showed that for strongly non-uniform attenuation and adequate statistics, very accurate attenuation, scatter and random corrections are needed to achieve accuracy better than 20% for most voxels, except close to regions with high gradient, where larger differences are seen. Insufficient scatter and random corrections lead to underestimation of the TC in regions with low attenuation. In such cases, tumors in the lung and close to the end of the axial FOV and with low statistics may not be seen. Reliable estimates of the quantification inaccuracy of 3D PET are needed if 3D PET is to be used for tumor segmentation and dose painting for radiation therapy.