Low-statistics reconstruction with AB-EMML

Abstract

In dynamic SPECT studies with short acquisition times per time-frame, data with very low-statistics is obtained. For such cases standard iterative reconstruction algorithms based on multiplicative correction factors, automatically including a non-negativity constraint, might not be Ideal. The AB-EMML algorithm allows the user to include prior information on the upper and lower bounds for the image values. We have used this algorithm with a negative lower bound for reconstruction of low-statistics SPECT data in order to allow for negative image values. Our results show that this method can preserve quantitative accuracy at low count levels, where standard methods produces biased values. Furthermore, the noise is much more uniformly distributed-lower in high intensity regions and higher in low intensity regions. The convergence is generally slower, but faster in cold regions.