On the Optimum Strategy for Subsets Based Reconstruction of In-Beam PET Data

At the heavy ion therapy facility at the Gesellschaft für Schwerionenforschung Darmstadt, Germany, an in-beam PET scanner is operated for quality assurance monitoring simultaneously to the therapeutic irradiation. The PET scanner, which is completely integrated into the treatment facility, registers the annihilation γ - rays following the decay of minor amounts of β+ radioactive nuclei produced via nuclear reactions between the ions of the therapeutic beam and the atomic nuclei of the irradiated tissue. From a comparison of the reconstructed activity distributions with those predicted from the treatment plan, deviations between the prescribed and the applied dose distributions can be detected. The solution currently implemented for the reconstruction of in-beam PET data consists of ML-EM and recently developed randomly filled subsets expectation maximization (RFS-EM) algorithms adapted to the low statistics case and to the dual head geometry. We proposed the reconstruction scheme for the subsets based RFS-EM algorithm. The scheme includes 4 iterations only, the number of subsets decreases from iteration to iteration and equals to 8, 6, 4, and 2 subsets, respectively. The scheme has been tested for two typical irradiation cases: head and neck (image space dimensions: 190 × 122 × 122 mm3) and pelvic (398 × 152 × 136 mm3) fields. The image quality was evaluated by means of visual inspection, root mean square error, and sensitivity of the method to the range deviations. The images reconstructed with the proposed RFS-EM scheme are of similar quality as those reconstructed with the reference 50 iterations of ML-EM. High quality of the system matrix is required for RFS-EM reconstructions of large fields.