Abstract
578
Objectives We describe a new approach to correct for effects of resolution blurring on activity estimates in dual-modality imaging. The higher resolution images (e.g., CT) are segmented into a few tissue types only within a small volume (VOI) surrounding an object of interest, e.g., a tumor.
Methods To estimate tissue activity concentration, measured emission projections are fitted to the sum of resolution-blurred projections of each tissue segmented within the VOI, scaled by its unknown activity concentration, plus a global background contribution obtained by reprojection through the reconstructed image volume outside the VOI. The approach was evaluated using multi-pinhole microSPECT emission data (~1.2mm res.) simulated for the MOBY mouse phantom containing two spherical lung tumors and one liver tumor (all 1.6mm diam.); each VOI was 4.8mm (12 voxels) cubed and, depending on location, contained up to 4 tissues (tumor, lung, liver, and heart) with relative concentrations 139:8:176;32. The modeled data correspond to a 98 MBq injection of a new Tc-99m agent and a 1-h acquisition. Since the parameters are estimated by linear maximum-likelihood fitting of projections, where the noise is Poisson and uncorrelated, convergence is guaranteed. Furthermore, mean parameter biases should be well approximated by estimates from noise-free data, while variances are given by the Cramer-Rao bound.
Results Tumor activity estimates all converged to ≤1% bias after an average of ~15 OSEM iterations, while the bias of activity-concentration in other VOI tissues ranged from -6.5% to +11.5%. Estimates from tissue regions near a high-contrast boundary (e.g., liver-lung) may be influenced by the Gibbs effect. The precision of tumor activity estimates ranged from 3.5% to 7.7%, while that of other tissues ranged from 0.8% to 13.8%.
Conclusions Projection-based fitting for tumor and nearby tissue activity is an accurate and robust approach to correct for effects of finite resolution in emission tomography.
Research Support Work supported by NIH R01 grants EB001989 and EB000802