Ing. Kateřina Macková

Introduction: The quality of fluorodeoxyglucose positron emission tomography (FDG-PET) images is limited by low effective resolution, which leads to blurring. The resolution can be additionally improved by partial volume correction (PVC) algorithms. The key input parameter influencing PVC performance is the correct specification of full width at half maximum (FWHM) of the scanner point spread function (PSF). The FWHM is standardly measured using a point source phantom under the standard protocol defined by the National Electrical Manufacturers Association (NEMA). However, the PSF varies in real conditions of human/animal neuroimaging and must be estimated individually. Therefore, a new method for fully automatic estimation of FWHM from FDG-PET images of the brain was proposed without the need for information on the scanner and protocol used. Methods: The method was based on estimating the parameters of edge spread function at the border of the gray matter from the FDG-PET image intensity function. The FWHM estimation algorithm was applied to FDG-PET brain images of 31 candidates for surgical epilepsy treatment in Motol University Hospital. The estimates were compared with FWHM measured according to NEMA standard on the same PET-CT scanners (Siemens Biograph mCT: 16 patients, Siemens Biograph Vision: 15 patients). Results: The median (interquartile range) value of estimated FWHM was 4 (3.9 - 4.1) mm, which was 11% lower than measured 4.5 mm for Siemens Biograph mCT scanner in transversal direction, and 6% lower 3.4 (3.3 -3.6) mm vs. measured 3.2 mm for Biograph Vision scanner. In the axial direction, estimated value 4.7 (4.5 - 5.2) mm was the same as the measured 4.7 mm for the Siemens Biograph mCT scanner, and 6% higher 3.6 (3.4 -3.8) mm vs. measured 3.4 mm for the Siemens Biograph Vision scanner. Conclusion: The proposed method can fully automatically estimate FWHM of PET scanner PSF only based on FDG-PET image. The estimates agreed with FWHM measured according to NEMA standard. FWHM estimation improves the performance of PVC algorithms that increase the resolution of FDG-PET images for the glucose metabolism assessment.