ADMM-EM Method for -Norm Regularized Weighted Least Squares PET Reconstruction

Figure 2

MAE values generated by the greedy enumeration strategy: (a) PWLS-EM with 400 iterations and (b) ADMM-EM with 400 outer iterations and 120 inner iterations. The global minimum MAE values correspond to the optimal penalty parameters. Note that both the - and -axes are log-scale to provide a clear visualization. PWLS-EM obtains the minimal MAE value (41.91) at , and ADMM-EM obtains the minimal value (13.05) at .

(a)

(b)