Application of the 3D row action maximum likelihood algorithm to clinical PET imaging

True 3D reconstructions from fully 3D PET data can yield high-quality images but at a high computational cost. The 3D row action maximum likelihood algorithm (3D RAMLA) with 3D spherically-symmetric basis functions (blobs) has recently been modified to reconstruct multi-slice 2D PET data after Fourier rebinning (FORE) but still using 3D basis functions (2.5 D RAMLA). In this study both 2.5 D RAMLA and 3D RAMLA were applied to different patient and phantom PET data to assess their clinical performance. Whole-body scans acquired on the C-PET scanner were reconstructed with FORE+FBP, FORE+OSEM, and FORE+2.5 D RAMLA for various reconstruction parameters (blob radius and shape, relaxation parameter). The 3D Hoffman brain phantom scanned on the HEAD Penn-PET scanner was reconstructed with 3DRP and 3D RAMLA, as well as FORE+OSEM. The authors' results demonstrate improvement of the RAMLA compared to the popular reconstruction methods in terms of contrast recovery and noise, especially in regions of limited statistics.