Relations Between System Matrices And The Complete Data Space In Mlem Using The Kullback-Leibler Distance

We present a quantitative method for relating system matrices to the complete-data space in maximum likelihood expectation maximization (MLEM) using the Kullback-Leibler distance. We show that a more accurate system matrix has a smaller Kullback-Leibler (KL) distance. System matrices of a coded aperture imaging system were used for comparison. The calculation of the KL distance is based on the Monte Carlo integral. We note that system matrices for the KL distance evaluation should be generated by underlying physics processes.