Skip to Main Content
Simultaneous dual isotope imaging (Tc-99m/I-123) has potential clinical applications but has not been implemented in the clinic. The aim of this work was to optimize the design of an artificial neural network (ANN) for cross-talk and scatter correction using a smaller number of energy windows (8) than we had previously proposed (26) to allow implementation on some clinical cameras, and to validate our approach using realistic Monte Carlo simulations and anthropomorphic brain phantom acquisitions. Monte Carlo simulations of dual isotope SPECT studies of a digital brain phantom and physical acquisitions of the striatal brain phantom were used to validate our approach. Corrected projections were reconstructed using an iterative OSEM algorithm that modeled non-uniform attenuation and variable collimator response in the projector/backprojector. In simulated I-123 images, ANN26 and ANN8 yielded similarly accurate (bias <7%) and similar results in all brain structures. An asymmetric windowing method (AW) yielded accurate estimation in the striata (bias <7%) but not in other brain structures. The estimation bias of Tc-99m primary activity was <10% in all brain structures with ANN26 and ANN8. This bias was greater than 25% in all brain structures with AW. In physical acquisitions, ANN26 and ANN8 yielded accurate estimation of I-123 activity in striata of both normal and reduced activity concentration (bias <7%). Bias significantly increased with AW between normal and reduced uptakes from 5 to 9%. The bias of the Tc-99m activity estimate was less than 6% with ANN26 and ANN8 but greater than 20% with AW. It is concluded that ANN8, which can be implemented more easily in the clinic than ANN26, yielded estimation bias and precision comparable to those of ANN26 in Monte Carlo simulations and physical acquisitions.