A new correction method for cross-talk using artificial neural networks: validation in simultaneous technetium and iodine cerebral imaging

Dual isotope-imaging has a high clinical potential in the assessment of brain perfusion (HMPAO-Tc-99m) and neurotransmission (IBZM-I-123) but is affected by cross-talk between the two radionuclides. We developed a new cross-talk correction method based on artificial neural networks (ANN). The ANN learning phase was performed using a Monte Carlo (MC) simulation of a geometrical phantom. The ANN correction was evaluated using independent MC simulations of a population of 15 normal and pathologic dual isotope studies of Tc-99m and I-123 based on a digital human brain phantom. Results from ANN were compared to those obtained with an energy windowing method (WI). All projections were reconstructed using filtered backprojection and compared to the unscattered activity distribution. Cross-talk in the I-123 image caused an overestimation of primary activity in the striata of 5% with ANN as compared to 14% with WI (p<0.01). In the Tc-9m image, cross-talk led to an error in the cortex smaller than 7% with ANN as compared to 40% with WI (p<0.01). Mean percent bias was less than 10% in all structures for both radionuclides with ANN. In this study ANN clearly outperformed WI and demonstrated to be a promising cross-talk correction method.