Fast Statistical Image Reconstruction for Emission Tomography: Application to SPECT
Nuclear medicine imaging, including PET and SPECT, has become a powerful diagnostic tool, in that nuclear medicine imaging is able to visualize physiological functions or functional metabolism. In nuclear medicine, an estimate of the three-dimensional spatial distribution of injected radionuclide is needed for diagnosis. This estimate is obtained through a reconstruction algorithm. Statistical image reconstructions have been proven to outperform the traditional FBP reconstruction in many aspects. However, the main problem with statistical reconstruction is its computation load and slow convergence. The OSEM (ordered subset expectation maximization) algorithm, which has an order-of-magnitude speed enhancement over the original MLEM (maximum likelihood EM) algorithm, is the most popular statistical reconstruction method used in many clinical hospitals worldwide. Nevertheless, the OSEM algorithm does not converge for real data so that its noise and resolution are difficult to predict. Furthermore, the OSEM is not easy to include a prior term. Many fast OS-type algorithms were proposed to solve for the speed and the convergence problems of the OSEM. They all need a user-specified relaxation schedule to ensure the speed and convergence property. But there is no easy way to decide the relaxation schedule. Previously, we had proposed a fast, convergent OS-type algorithm, called COSEM-MAP, to retain the speed while at the same time without any relaxation schedule. Here, we briefly review the COSEM algorithm, and then apply the method to a set of real SPECT phantom data acquired from the Chang Gung Memorial Hospital. A segmented attenuation correction method is applied to the SPECT data. The COSEM-MAP result is compared to those of the popular FBP and OSEM methods often used in clinical reconstruction. Our COSEM-MAP method shows potential effectiveness as compared to the FBPand OSEM.