Accelerated time-of-flight (TOF) PET image reconstruction using TOF bin subsetization

Objectives Time-of-flight (TOF) PET has regained popularity for improving image quality and lesion detectability in clinical PET examinations. Using TOF technology, the spatial location of annihilation events is estimated and incorporated into image reconstruction, thus leading to reduced noise propagation and improved convergence rate. The ordered subsets expectation maximization (OSEM) is the standard PET image reconstruction algorithm used in clinical setting owing to its fast convergence compared to maximum likelihood expectation maximization (MLEM). In this study, we propose a novel approach to further improve the convergence of TOF-OSEM algorithm through subsetization of emission data over TOF bins as well as azimuthal bins.

Methods In the proposed approach, TOF PET data are subsetized by interleaving TOF bins based on the number of TOF subsets, thereby PET images were updated over interleaved segments of response, leading to reduced inter-voxel dependencies and thus improved convergence. The contrast recovery coefficient (CRC) versus image roughness (IR) performance of both OSEM and MLEM algorithms with and without TOF subsetization was evaluated using experimental NEMA phantom and clinical FDG PET/CT studies acquired on the Siemens mCT PET/CT scanner. In this first study, the combination of 14 azimuthal subsets, and 2-3 TOF subsets were evaluated.

Results The proposed technique improved considerably the convergence of OSEM and MLEM algorithms. For the NEMA phantom, the OSEM algorithm (with 14 azimuthal subsets) resulted in overall CRC and IR of 65.8% and 21.2% over all spheres, respectively, while the accelerated OSEM (with 14 azimuthal and 3 TOF subsets) resulted in overall CRC and IR of 70.0% and 31.2%. The clinical study also demonstrated that the proposed method further accelerated the convergence, thus providing a high lesion-to-background ratio after a fewer number of iterations.

Conclusions TOF subsetization is a promising technique to further improve the convergence properties of OSEM and MLEM algorithms.

Research Support This work was supported in part by the Swiss National Science Foundation under Grant SNSF 31003A-149957 and by the Indo-Swiss Joint Research Programme ISJRP 138866.