Comparison of statistical reconstructions with isotropic and anisotropic resolution in PET

Statistical reconstruction methods based on the penalized maximum likelihood (or maximum a posteriori) principle have gained increasing attention in emission tomography. Fessler and Rogers have shown in 1996 that penalized maximum likelihood reconstruction with a conventional quadratic penalty results in anisotropic point spread functions (PSFs). Since then several approaches have been developed to design modified penalty functions to achieve isotropic PSFs. While an image with an isotropic PSF may be useful in some situations, its performance on clinical detection and quantitation tasks is unknown. In this paper we compare the task performances between reconstructions with isotropic and anisotropic PSFs using computer simulations. The performance on lesion detection is measured by a channelized Hotelling observer, and the performance on region of interest quantitation is evaluated by the bias and variance tradeoff. The results show that reconstructions with a conventional quadratic penalty function (anisotropic resolution) outperform post-smoothed maximum likelihood reconstructions (isotropic resolution) for both tasks, which indicates that isotropic resolution may not be suitable for lesion detection and quantitation.