Integrating Anatomical Priors in ECT Reconstruction via Joint Mixtures and Mutual Information

We present a joint mixture model for integrating anatomical information in ECT reconstruction. The mixture model acts as a Bayesian prior with the added benefit of the anatomical segmentation pixel-region labels guiding the reconstruction process. An additional hyper-prior on the mixture occupation probability is introduced. This has the novel interpretation of maximizing the mutual information between the anatomical image and the evolving ECT reconstruction. Results are presented on a phantom with the bias/variance tradeoff being the indicator of performance.