Regularized versus non-regularized statistical reconstruction techniques

Abstract

Abstract An important feature of positron emission tomography (PET) and single photon emission computer tomography (SPECT) is the stochastic property of real clinical data. Statistical algorithms such as ordered subset-expectation maximization (OSEM) and maximum a posteriori (MAP) are a direct consequence of the stochastic nature of the data. The principal difference between these two algorithms is that OSEM is a non-regularized approach, while the MAP is a regularized algorithm. From the theoretical point of view, reconstruction problems belong to the class of ill-posed problems and should be considered using regularization. Regularization introduces an additional unknown regularization parameter into the reconstruction procedure as compared with non-regularized algorithms. However, a comparison of non-regularized OSEM and regularized MAP algorithms with fixed regularization parameters has shown very minor difference between reconstructions. This problem is analyzed in the present paper. To improve the reconstruction quality, a method of local regularization is proposed based on the spatially adaptive regularization parameter. The MAP algorithm with local regularization was tested in reconstruction of the Hoffman brain phantom.