Computationally efficient iterative transmission imaging for the Inveon DPET

Transmission measurements are performed on PET tomographs to create attenuation correction factors in support of quantitatively accurate images. The Inveon dedicated PET (DPET), a state-of-the-art small animal imaging system is an example of such a system. We report on the design and implementation of an iterative reconstruction approach that incorporates an explicit model of the system geometry. Although iterative algorithms are known to be computationally costly compared with analytic reconstruction methods, we show that the weighted least-squares computation considered in our case can be made computationally efficient. Indeed, we show that a high-quality transmission image can be reconstructed in under a minute. Computational speed is gained algorithmically through the use of relaxation and ordered-subsets, and implementation wise through the use of Intel´s SSE vector instructions and multi-core techniques. Experimental results are provided for both phantom and real data.