Evaluation of filter function for volume PET imaging using the 3DRP algorithm

We have investigated the influence of filter and its cutoff frequency on the image quality for volume PET imaging using the widely used 3D-reprojection (3DRP) algorithm. An important parameter in 3DRP and other filtered backprojection algorithms is the choice of the filter window function. For this work, three different low-pass filter window functions, Hann, Hamming and Butterworth, were investigated. For each filter a range of cutoff frequencies were considered. Projection data were acquired by scanning a uniform cylindrical phantom, a cylindrical phantom with four small lesions and the Hoffman brain phantom. All measurements were performed with the high-resolution PET camera developed at MD Anderson Cancer Center (MDAPET). This prototype camera, which is a multiring scanner with no septa, has a transaxial resolution of 2.8 mm. The evaluation was performed by computing the noise level of reconstructed images of the uniform phantom, the contrast recovery of the hot lesions in warm background, and by visual inspection of image quality for the Hoffman brain phantom. For the high statistics data presented here, a cutoff frequency 0.6 to 0.8 of Nyquist resulted in a reasonable compromise between the contrast recovery and the noise level for the Hann filter. For the Butterworth filter, a cutoff at 0.4-0.6 of Nyquist frequency was a reasonable choice. Overall, the Butterworth filter performed better in contrast recovery and spatial resolution at the cost of somewhat noisier image.