Influence of TOF resolution on object dependent convergence in iterative listmode MLEM

Abstract

167

Objectives: Because MLEM is shift variant same size lesions with equal contrast will converge differently depending on the location in the object and the surrounding activity. In Time-of-flight (TOF) PET the shorter point spread function limits the influence of objects further away. Therefore the convergence is expected to be less dependent on the object. The goal of this study is to determine improvements in convergence and quantification as a function of timing resolution.

Methods: The Gate Monte Carlo simulator was used to generate listmode data (with different TOF resolutions of 100, 200, 400, 800 and 1600 ps) of two phantoms. The geometry was based on an existing prototype TOF-PET system. No scatter or attenuation in the phantoms were simulated. The first phantom contained two hot spheres (2.5 mm radius) at equal distances from the center. The right hotsphere is in a cold region. Above and below the left hotsphere are larger spheres (5 cm radius) which reduces the convergence rate of the hotsphere. The second distribution was a slice (through the heart) of the Mcat phantom. Seven spheres with a contrast of 2:1 were introduced at different locations in low activity regions like the lung and high activity regions like the body. The listmode data were reconstructed with a 3D TOF listmode MLEM algorithm.

Results: The free point source converges after about 20-30 iterations. For non-TOF data (1600 ps) a high number of iterations (several hundred) are needed for convergence of the enclosed hotsphere. When timing resolution improves, less influence of the surrounding spheres is seen on the convergence rate of the enclosed sphere. The contrast was determined at 100 iterations. The ratio between the contrast of the enclosed and the free hotsphere improved from 58 % for 1600 ps to 72 % for 400 ps and 98 % for 100 ps. In the Mcat phantom there is a quite large variation on the contrast of the seven spheres (from 1.46 to 2.03) for the 1600 ps data. When timing resolution improves the variation on the contrast is reduced. For example at 100 ps the contrast of the seven spheres lies in the range of 1.81-1.91 with a standard deviation of 7%, whereas the st.dev. increases to 10%, 15%, and 19% at 200ps, 400 ps,and 800 ps, respectively.

Conclusions: The difference between the convergence of the different spheres reduces clearly when better timing resolution is available. TOF PET will result in faster convergence that is less influenced by the object. Therefore more accurate quantification will become possible with TOF-PET.