Reduction of variance in multispectral pet data by energy space filtering

Statistical fluctuations due to the use of multiple energy windows and short acquisition times limit the potential of multispectral positron emission tomography (PET) as a more flexible environment for data processing. This is mainly because such fluctuations make normalization for detector efficiency and random correction inefficient. Smoothing data in the energy space is proposed as an approach to reduce statistical variance in the projection space without degrading resolution. Smoothing 2.5 minute acquisition data from a uniform source is shown to produce distributions in both energy and projection space similar to those obtained for 12 hours. Irrespective of counting statistics, the normalization in conjunction with energy space filtering reduces the overall variance by similar amount. It is concluded that the systematic variance in projection space and statistical variance in energy space can be reduced independently and for best result, they should be used together.