Gaussian and point spread function filtering on a Philips Ingenuity TF-128 PET/CT scanner

Objectives Quantitative accuracy in multicenter FDG studies can be improved by harmonized imaging procedures. In Europe EARL accredits PET centers for multicenter studies and is widely adopted in large image trials. Recovery coefficients (RC) should be within specified limits. Newer generation scanners often have higher RC. In this study the use of a Gaussian filter (in line with EARL accreditation) is shown as well as the effect of post-acquisition point spread function (psf) processing on test retest (TRT) variability of lesion quantification.

Methods A NEMA-NU 2001 image quality phantom was used to derive RC max and background variability. Six measurements were performed on a Philips Ingenuity TF-128 PET/CT scanner (Philips Healthcare, Best, The Netherlands) according to EARL guidelines. Data were reconstructed and analyzed without additional filtering (std), after filtering with a 4 mm FWHM Gaussian (earl) and after processing with recommended psf settings (psf). The same three options were applied to 5 test retest FDG whole body studies (NSCLC) scanned within one week on a Gemini TF-64 PET/CT scanner. Tumor to liver ratios (TLR) based on SUV max of the primary tumor and SUV mean of the liver (cf PERCIST) and their TRT variability were derived.

Results Figure 1 shows RC_max values as a function of sphere volume. Average coefficient of variation (CoV) in the background was 11.5% (std), 10.5% (earl) and 13.3% (psf) for the three filtering options, respectively. Average TRT variability for the TLR was comparable for all three options (range 1-30%, 12% on average). Average SUV max of the tumor was 23% and 4% higher for the psf and std options, respectively, compared to the earl option.

Conclusions TRT performance was not affected by psf filtering. However, introduction of psf leads to significantly higher observed SUVs, and should therefore not be introduced during ongoing (multicenter) studies.