A generalized approach to scatter correction using multiple energy windows

The authors studied the performance of methods for scatter correction in Tc-99m and Tl-201 imaging, using a Rollo phantom and a gamma camera modified to record (xyE) in list mode form. They compared the performance of these methods to optimal linear methods and to locally adaptive linear methods, and to the limits arising from the presence of Poisson noise. For Tc-99m, only modest improvement over the performance of dual-window methods is possible with any linear method. At typical clinical count rates, mean square error is dominated by Poisson noise (and so cannot be improved by any method), although a slight reduction in bias may be possible with locally adaptive techniques or by smoothing the scatter estimate. Dual- or triple-window techniques are further from optimal for Tl-201, and locally adaptive techniques, such as spectral fitting, may be more useful