Abstract
Simultaneous Tc-99m/I-123 dual-isotope SPECT imaging allows assessment of two physiological functions under identical conditions, without any image registration. However, the separation of these radionuclides is difficult, because their energy is close. Images can be severely distorted due to crosstalk. Use of an artificial neural network (ANN) has been previously shown to be an effective tool in compensating crosstalk and scatter. Conventional ANN techniques require a large number of energy windows (>8); however, such a capability is not available in most clinical SPECT systems. In this study, we proposed a filter method to improve the ANN techniques with only 4 energy windows. In this work, we chose 0.2 mm gold filter to separate Tc-99m and I-123, based on the attenuation difference of the gamma of each isotope. The filter is placed over the collimator. We designed an ANN with 8 input, 10 nodes in the hidden layer, and four nodes in the output layer. The inputs were count ratios in four energy windows from scans with and w/o filter. The outputs layer provided the ratio of estimated primary to total photons for Tc-99m and I-123, w/ and w/o filter. The outputs of the ANN are then combined to form two primary data with crosstalk and scatter corrected each for the Tc-99m and I-123. A GATE/MPHG Monte Carlo code is used for the DISA SPECT simulation. In this work, a three-rod phantom and a NCAT phantom were used to validate the proposed method. The results showed that quantitative recovery from dual isotope image was comparable to that from single-isotope imaging. Compared to images using other crosstalk compensation methods, our results were much better than that of asymmetric window method and closed to the results obtained from conventional ANN technique. Using the filter method incorporated with ANN, crosstalk was successfully compensated and only 4 energy window were needed.