Abstract
355
Objectives PET Standardized Uptake Values (SUVs) are highly dependent on image reconstruction protocols, e.g., number of iterations, smoothing, etc. This can preclude quantitative comparison between images reconstructed with different protocols. We present an approach that enables quantification of a reference SUV (SUVref) that is less dependent on reconstruction and facilitates comparison between differently reconstructed images.
Methods To compute SUVref for a given reconstruction, the image volume is convolved with an additional 3D Gaussian filter, selected to reduce inter-reconstruction variation in SUV. A region of interest (ROI) created on the original unfiltered image is propagated to the hidden filtered image and the maximum SUV within that ROI returned as SUVref. The size of the filter applied is computed from an analysis of an acquired NEMA Image Quality phantom, reconstructed with the same protocol as used for the clinical dataset. The filter selected is that producing measured sphere-to-background ratios (SBRs) for each of the phantom hot spheres closest to a predefined set of SBRs, derived from an analysis of a diverse set of 11 reconstructions. Phantom-optimised filters were applied to 10 whole body clinical datasets acquired on a Siemens Biograph mCT and reconstructed with 4 different protocols (including 3D-OSEM, PSF and TOF) and the variation in SUVmax and SUVref due to differences in reconstruction alone were compared.
Results The phantom-optimised filters reduce the mean percentage difference in SUV across the 4 reconstruction protocols from 17.9% (SD: 17.4) to 1.6% (SD: 9.8) with SUVref, over a set of 50 ROIs in each of the 10 datasets. Furthermore, a percentage change in SUV of greater than 30% is observed in 19.9% of cases (i.e., same ROI, different reconstructions) evaluated using SUVmax compared to only 1.0% using SUVref.
Conclusions SUVref substantially reduces reconstruction-dependent variation in SUV measurements, potentially increasing confidence in quantitative comparison of clinical images for monitoring treatment response or disease progression