A prior image model with mixed continuity constraints for Bayesian PET image reconstruction

Describes a novel image prior model with mixed continuity constraints for Bayesian PET image reconstruction. If an image can be partitioned into partial-volume and various pure regions, the author models the image as a piece-wise smooth function through a Gibbs prior. Within each pure or partial-volume region, the image intensity is governed by a thin-plate energy function. Both first- and second-order edge detection techniques are applied to estimate region boundaries. Instead of using the binary processes representing region boundaries, a controlled-continuity approach is adopted to influence boundary formation. The rationale is that while the first-order edge detection captures the jumps between two pure regions, the second-order one locates the crease connecting a pure region to partial-volume region. This model is referred as a modified weak-plate model with controlled continuity. Results indicate that a superior improvement of image quality using this approach can be obtained.