Quantitative I-131 SPECT reconstruction using CT side information from hybrid imaging

Abstract

A penalized-likelihood (PL) SPECT reconstruction method using a modified regularizer that accounts for anatomical boundary side information was implemented to achieve accurate estimates of both the total target activity and the activity distribution within targets. In both simulations and experimental I-131 phantom studies, reconstructions from 1) penalized likelihood employing CT-side information based regularization (PL-CT); 2) penalized likelihood with edge preserving regularization (no CT); 3) penalized likelihood with conventional spatially invariant quadratic regularization (no CT) were compared with 4) Ordered Subset Expectation Maximization (OSEM), which is the iterative algorithm conventionally used in clinics for quantitative SPECT. Evaluations included phantom studies with perfect and imperfect (misregistered) side information and studies with uniform and non-uniform activity distributions in the target. For targets with uniform activity, the PL-CT images and profiles were closest to the ‘truth’, avoided the edge offshoots evident with OSEM and minimized the blurring across boundaries evident with regularization without CT information. Apart from visual comparison, reconstruction accuracy was evaluated using the bias and standard deviation (STD) of the total target activity estimate and the root mean square error (RMSE) of the activity distribution within the target. PL-CT reconstruction reduced both bias and RMSE compared with regularization without side information. When compared with unregularized OSEM, PL-CT reduced RMSE and STD while bias was comparable. For targets with nonuniform activity, these improvements with PL-CT were observed only when the change in activity was matched by a change in the anatomical image and the corresponding inner boundary was also used to control the regularization. In summary, the present work demonstrates the potential of using CT side information to obtain improved estimates of the activity distribution in targets without sacrificing the accuracy of total target activity estimation.