A comparative study of the effects of using higher order mechanical priors in SPECT reconstruction

While the ML-EM reconstruction method in SPECT is unstable due to its ill-posed nature, the Bayesian reconstruction methods overcome this instability by introducing the prior information. The new prior, the weak plate, is attractive in that it models spatial structures more accurately than various forms of smoothness priors used in the past. Here the authors argue that more expressive priors able to model more complicated forms than a piecewise constant source may be better used in SPECT reconstructions. To evaluate quantitative performance of the reconstruction algorithms, in an ensemble sense, bias and variance were used as a metric. The authors show that the extension to a higher-order model significantly improves variance at approximately no cost in bias error. The authors also observe and characterize the behavior of the associated hyperparameters of the prior distributions in a systematic way