Residual correction algorithms for statistical image reconstruction in positron emission tomography

Positron emission tomography (PET) is a radionuclide imaging modality that plays important roles in visualizing, targeting, and quantifying functional processes in vivo. High-resolution and quantitative PET images are reconstructed by solving large-scale inverse problems with iterative algorithms that incorporate accurate physical and noise modeling of the imaging process. The computation demands of PET image reconstruction are rapidly increasing as higher-resolution detectors, larger imaging field-of-view, and dynamic or adaptive data acquisition modes are being adopted by modern PET scanners. The trend of the increase in the computation demands is even faster than Moore's law that describes the exponential growth in the number of transistors placed on an integrated circuit.;In this project a residual correction mechanism is introduced to PET image reconstruction to create computationally efficient yet accurate tomographic reconstruction algorithms. By using residual correction, reconstruction methods are able to adopt a more simplified physical model for fast computation while retaining the accuracy of the final solution. Residual correction can accelerate existing image reconstruction packages. It allows iterative reconstruction with more accurate physical models which are currently impractical due to their high computation cost. Two illustrative applications of the residual correction approach are provided. One is image reconstruction with an object-dependent Monte Carlo based physical model. The other is image reconstruction using an ultra fast GPU-accelerated simplified geometric model.