Iterative reconstruction of SPECT data with adaptive regularization

A nonlinear regularizing least-square reconstruction criterion is proposed for simultaneously estimating a single-photon emission computed tomography (SPECT) emission distribution corrected for attenuation together with its degree of regularization. Only a regularization trend has to be defined and tuned once for all on a reference study. Given this regularization trend, the precise regularization weight, which is usually fixed a priori, is automatically computed for each data set to adapt to the noise content of the data. We demonstrate that this adaptive process yields better results when the noise conditions change than when the regularization weight is kept constant. This adaptation is illustrated on simulated cardiac data for noise variations due to changes in the acquisition duration, background intensity, and attenuation map.