The effect of PET reconstruction algorithms on tumor delineation for radiation oncologists

Objectives Partial volume effects in PET can lead to over- or under- estimation of tumor volumes depending on the local activity. This work investigates the extent of over- or under-estimation.

Methods A Jaszczak phantom with hollow spheres of various sizes (0.9mm - 31mm diameter) was filled with F-18 water using 3 different sphere-to background ratios (SBR), ranging from 3.5:1 to 10:1. Background was 0.07mCi/Kg approximating the average activity in a radiation oncology patient at the time of imaging. For each SBR, a single acquisition PET/CT was performed and the RAW data were reconstructed by using OSEM with 32 subsets with 2 iterations, then varying the FWHM of the Gaussian Blurring filter from 5 - 8 mm, and using image matrix sizes of 128 X 128, 192 X 192 and 256 X 256. Regions were automatically segmented based on a clinically used SBR threshold of 2.5, and then the threshold was varied to be about 50% of the actual SBR. The volumes as determined by the software were compared to the volumes as derived from the CT. SBR values were also compared to measured values for all sphere sizes.

Results For all SBR of 3.5 the ratio of PET to CT volumes ranged from 0.92 down to 0.0 for the smallest sphere, with a sharp drop occurring for spheres < 19mm. The 8 mm blurring filter had the most dramatic the drop off for the PET volumes of these smaller spheres however the image matrix size had more effect for the smallest sphere. Using a filter FWHM of 5 mm with 256 X 256 matrix, and 50% SBR yielded the most accurate volume determination for SBR 10 with PET/CT volumes of 1.15 - 1.0 for all spheres. The same reconstruction parameters also yielded the best ratios for the SBR 3.5 data, where the PET/CT ratio ranged from 1.14 for the largest sphere down to 0.3 for the smallest.

Conclusions The apparent size of a lesion is dependent on SBR and reconstruction technique. A FWHM Gaussian filter provided the most accurate results overall. But a radiation oncologist should not segment tumor based on PET images alone