Quantitative image reconstruction in simultaneous 123I/ 99mTc myocardial SPECT

In the ¹²³I/^99mTc myocardial SPECT study the image reconstructed with the counts of primary photons emitted from the low photopeak energy (^99mTc) radionuclide is distorted by Compton scattered photons originating in the high photopeak energy (¹²³I) radionuclide. To correct the scattered photons included in the low energy photopeak window in the simultaneous data acquisition, we proposed a method with a neural network at the IEEE MIC in 1999. This method uses a layered neural network (input units: 10, hidden units: 20 and output units: 2). El Fakhri et al. (2001) also proposed a method at the IEEE MIC in 1999 which was like our approach, and they obtained good results. The major differences between their method and ours are (1) the energy range for data acquisition, (2) the width of the narrow energy window, and (3) the method for calculating the number of primary photons. This paper investigates the performance of their method and ours in consideration of the above three points. Performance was evaluated with Monte Carlo simulation data and experiment data. In the simulation we used the MCAT phantom and accuracy was evaluated by the mean squared error in the reconstructed images. The results indicated that the accuracy of our method was slightly superior to El Fakhri´s method. In the experiment we fixed the energy range to 120-180 keV and changed the width of the narrow energy window to 2.2, 4.4 and 6.7 keV. The results showed that a window width of 2.2 keV was too narrow for as an energy window. In conclusion, it was confirmed that our method was slightly superior to El Fakhri´s method and both methods were fairly effective in separating ^99mTc and ¹²³I accurately