Performance Characterization of a Feature-Matching Axial Smoothing Method for Brain PET Images

A feature-matching axial smoothing method for images obtained with new generation positron emission tomography (PET) scanners has been introduced before and has been shown to enhance visually the signal-to-noise ratio without generating adverse artifacts. This chapter studies the performance of the method quantitatively by applying the method to multiple realizations of computer-simulated brain PET images and to PET images of the Hoffman brain phantom. The pixel-by-pixel noise level was calculated as the sample standard deviation of multiple noise realizations. Spatial resolution degradation was evaluated by the frequency response function of the smoothing procedure. For brain PET images, the results show that the feature-matching method causes a smaller spatial resolution degradation than the conventional axial smoothing method that uses the same smoothing filter. For the conventional method to give a comparable resolution loss, a shorter smoothing filter must be used, which would result in less noise reduction (equivalent to ~ 70% fewer photon counts than with the new method). Moreover, the amount of resolution degradation and/or noise advantage of the feature-matching method does not appear to be dependent on the in-plane resolution or noise level of the original images.