Stochastic Optimization For 3D ECT Data Correction And Image Reconstruction

Emission Computed Tomography (ECT) includes Positron Emission Tomography (PET) and Single Photon Emission Computed Tomography (SPECT), which makes use of radio isotopes labeled tracers being involved in the metabolism in vivo. ECT records the 7 photons emitted by the radio tracers, that provide the metabolic changes between the normal and abnormal organs. The quantitative accuracy of acquired data and reconstructed images by ECT is crucial for understanding the fundamental life processes and disease diagnosis.With the development of ECT, 3D acquisition mode improves the system sensitivity significantly, and has the potential to reduce the acquisition time. However, 3D acquisition mode introduces more scatter coincidence events, and more coincidence events from outside FOV at the same time; besides, the increase of number of detector blocks and detector efficiency, and the complicated characters of 3D data will limit the applications of 3D acquisition mode of ECT. To solve these problems, new data correction methods for 3D data and more suitable, robust reconstruction methods should be introduced.This paper introduces stochastic optimization methods for 3D ECT data correction and image reconstruction. The applications of stochastic optimization are mainly in two forms: 1. Optimal solution by Monte Carlo methods; 2. Optimal solution to deal with system noises. In this paper, the optimal solution by Monte Carlo methods is used for 3D data correction, and the optimal solution to deal with system noises is used for ECT image reconstruction by state-space framework. The main contributions of this paper are shown as follow.As the basis of data correction, accurate ECT system performance evaluation is important for the proper use of ECT systems. Based on the international standards for whole body PET performance evaluation, we performed system evaluation for the first animal PET scanner in the mainland of China, provided the evaluation methods and results of system spatial resolution, sensitivity, scatter fraction and image quality.After system evaluation, this paper modeled whole body PET scanner and animal PET scanner with tungsten septa respectively using Monte Carlo simulations, analyzed the scatter properties of both PET systems, including scatter fraction, scatter distribution, multiple scatter events, scatter events from outside FOV in 3D acquisition mode and scatter fraction, scatter distribution affected by tungsten septa in 2D acquisition mode.Making use of the scatter properties analysis, this paper provided a scatter correction method based on Single Scatter Simulation (SSS). To deal with two difficulties in scatter correction: multiple scatter events and scatter events from outside FOV, a new scale method was introduced. By this method, more accurate scatter correction was achieved. The software has already been applied on a real 3D whole body PET scanner.For image reconstruction, this paper extended existing state-space framework for more robust ECT image reconstruction, took the uncertainties in system probability matrix and acquired data into consideration, introduced a uncertainty penalized weighted least square framework for PET image reconstruction, which provided a new idea for dealing with the uncertainties in ECT acquisition system.Another extension for state-space framework is the simultaneous estimation of activity distribution and attenuation map, this paper made use of Unscented Kalman Filter (UKF) to reconstruct the activity distribution and attenuation map sequentially. At the same time, UKF can provide better results when dealing with system noises and system modeling errors.In order to promote the study of multi-tracers, this paper first established a dynamic PET reconstruction framework for activity distributions of multi-tracers. According to the theory of multi-injection, single-acquisition, parallel compartment models of multi-tracers was used as state equation in state-space framework, the composite photon counting procedure was used as measurement equation, so the ECT image reconstruction problem became a state estimation problem under state-space framework. At last, a robust H_∞filtering was used for system estimation. This method first finished the simultaneous estimation of activity distributions of multi-tracers.