Optimal locally adjustable filtering of PET images by a genetic algorithm

Images produced from positron emission tomography (PET) data sources are important for detecting tumors. To improve the resolution of images produced by PET scanners we develop a procedure (optimal locally adjustable filtering OLAF) where the parameters of filters (eg, Metz, Gauss) are adjusted "optimally" at each point of the reconstructed image in terms of a global objective function, which combines goodness-of-fit, and entropy terms The optimization problem is solved by a specialized genetic algorithm. The approach was tested on several PET data sets (simulated and clinical) (Levkovitz et al., 1998) and has demonstrated that OLAF improves both contrast and uniformity, compared to usual fixed post-processing filtering methods.