Methods for reducing statistical noise of multispectral data in positron emission tomography

The combination of multiple energy windows, short acquisition times and random correction inevitably amplifies statistical noise in multispectral PET data. With practical data, random subtraction amplifies noise to a point where this approach is not feasible. In such situations, the benefits of normalization may not be realized. Noise reduction in the energy space prior to random subtraction and normalization is therefore essential. In this work, we describe noise and random correction techniques which can be used in conjunction with normalization procedures to overcome this problem. Like normalization, noise suppression relies on scaling factors derived from measurements obtained with high statistics. The technique has been tested as a function of statistics using a flood source. We conclude that noise reduction in energy space prior to random correction and normalization can effectively minimize statistical fluctuations in multispectral data.