Monte Carlo Simulation of Single Photon Emission Computed Tomography.

A Monte Carlo code was used to simulate the scatter processes which occur in Single Photon Emission Computer Tomography. The transport of photons from their emission by the radioactive source within the phantom to their detection by the nuclear medicine camera was modeled. The simulation was tested by comparing qualitatively and quantitatively the theoretical with the experimental energy spectra. Also, experimental and theoretical scatter fractions and detector efficiencies were compared. In addition, two phenomenological correction algorithms, the Gaussian Subtraction and Least Squares Fit methods, were studied for their ability to correct for scatter. Eighteen cases involving phantoms containing uniform and nonuniform biological media along with a point or distributed source were studied. Cylindrical symmetry of the phantom and source was assumed in order to reduce the computer runtime. The theoretical energy spectra generated by the Monte Carlo code agreed with the experimental spectra to within one standard deviation. The exceptions involved regions in the energy spectrum below 120 keV for a radioactive source in air and the nonscattered peak region for the modeling of bone. The energy spectra were found to be dependent on the medium and radioactive source distribution within the phantom. The Gaussian Subtraction method was found to give a relatively constant mean difference of 4 +/- 4 percent between the "theoretical" and "estimated" scatter fractions for windows of up to 22 keV in width. For wider windows, the mean difference was found to increase linearly with width. The Least Squares Fit method was found to parallel the results given by the Gaussian Subtraction method for windows of up to 22 keV in width. The mean difference, though, was found to increase much less rapidly for wider windows that that found in the Gaussian Subtraction method thereby making this method more attractive. Both quantitative methods had difficulty with cases involving an effective linear attenuation differing from that most commonly found. Such cases had to be treated separately in the Least Squares Fit method but they could be treated along with the others in the Gaussian Subtraction method with the penalty of a larger error than usual.