Reduction of noise amplification in SPECT using smaller detector bin size

Abstract

In SPECT iterative reconstruction methods, such as the ML-EM (Maximum Likelihood Expectation Maximization) algorithm, the noise propagation from the projection measurements into the reconstructed image has been a difficult problem to control as the algorithm iterates. In this paper, we show that the noise amplification at high number of iterations can be reduced by using a detector whose bin size is smaller than the image pixel size without applying any regularization methods or changing any other factors. We compare different detector system characteristics using SVD (Singular Value Decomposition) analysis, show the noise properties in each detector system through both simulation studies and physical phantom studies, and finally compare how the noise amplification affects the image quality in different detector systems. The ML-EM algorithm when used in conjunction with a smaller detector bin size has better convergent properties and reduces noise amplification at high number of iterations.