PRACTICAL CONSIDERATIONS OF THE WIENER FILTERING TECHNIQUE ON PROJECTION DATA FOR PET

Wiener filtering (WF) attempts to recover a blurred object by incorpor ating the system modulation transfer function (MTF) and its noise char acteristics. Although it has been widely studied in SPECT, few studies have been reported in PET. In this investigation, first, we discuss s ome factors in PET which affect the Poisson nature of the noise on dat a, such as the axial normalization, detector efficiency correction and randoms correction. Then we propose a modified WF which compensates f or the above factors. The modified WF provides a more accurate noise e stimation method and a simple way to optimize the filter performance. Finally we apply the modified WF to projection data of the volumetric UGM PENN-PET 240H scanner in both transaxial and axial directions. A c old-hot sphere phantom and 3D brain phantom are used for evaluation. T he modified WF is compared to the conventional WFs. Our results indica te that 1) the conventional WFs may overestimate the noise power spect rum; 2) although the WFs are sensitive to the image statistics their o verall performances are still quantitatively much better than without any restoration filtering; 3) in general, filtering in both axial and transaxial directions is better than the transaxial filtering only, bu t its superiority is more obvious when applied to scans with low image statistics; 4) In general, the Wiener filtering improves the PET imag e contrast and count recovery.