Assessment of image quality with a fast fully 3D reconstruction algorithm

True three-dimensional (3D) reconstructions from fully 3D positron emission tomography (PET) data yield high-quality images but at a high computational cost. Image representation using three-dimensional spherically-symmetric basis functions on a body-centered cubic (BCC) grid, as opposed to a simple cubic (SC) grid, can reduce the computational demands of a 3D approach without compromising image quality by reducing the number of image elements to be reconstructed. The goal of this study was to determine if the image quality improvements predicted for the 3D row action maximum likelihood algorithm (RAMLA) over 2.5D RAMLA after Fourier rebinning (FORE) would be seen with clinical PET data. Torso phantom, whole-body patient, and brain patient studies were used in this analysis. Data were corrected for detector efficiency, scatter, and randoms prior to reconstruction. Attenuation effects were either incorporated into the system model or pre-corrected prior to reconstruction. Higher contrast at comparable noise levels (or lower noise for comparable contrast) are seen with 3D RAMLA (SC or BCC grid) for both phantom and patient data. The brain patient data show improved axial resolution with 3D RAMLA, where the degradation in resolution with FORE is eliminated. Application of a fully 3D reconstruction algorithm is possible in clinically reasonable times