Abstract
Algebraic reconstruction techniques (ARTs) have been intensively studied in image reconstruction. However, reconstructed images often exhibit a characteristic noise when applied to emission tomography. In this work, we study the influence of the ART relaxation parameter and identify values of that parameter which yield smooth images and good convergence. The study includes both simulated and real single photon emission computed tomographic (SPECT) data. In the simulated studies, scattering, attenuation, noise and detector response were included. Relaxation factors ranging between 0.01 and 0.35 were considered. Total counts in projections were between 
and 
in one slice. The goodness of the reconstructions was assessed by using the correlation coefficient between the pattern and the actual reconstruction (CC), the background coefficient of variation (CV), the contrast (CON) and the signal-to-noise ratio (SNR). The results show that values for the relaxation factor and number of iterations close to 0.1 and 8, respectively, yield reconstructed images in which CC, CV, CON and SNR have well balanced values and CC, CV and SNR are very close to the extremal. An increase in either the number of iterations or in the relaxation factor results in a rise of CON but the other three FOMs deteriorate. The results from the real phantom, which are in agreement with those of the simulated studies, validate the simulation method and demonstrate the influence of the relaxation parameter.