A penalized Algebraic Reconstruction Technique (pART) for PET image reconstruction

The complexity measure (CM) can be expressed as the number of basis functions, e.g. voxels or wavelet basis functions, for decomposing an image. It provides flexibility in describing a prior. A wavelet based complexity penalty term is used in this paper. This penalty stems from a multi-scale Poisson likelihood factorizations of the image formation. So it is capable of extracting the signal from Poisson noise polluted data. ART (algebraic reconstruction technique) is one of the most important iterative reconstruction algorithms. The pART (penalized Algebraic Reconstruction Technique) is proposed by using this regularization technique as a penalty term in the ART algorithm. First, the fact that ART implies an objective function is pointed out. The CM penalty term is then added in the objective function of ART. In this way, it is integrated into the iterative framework of the ART algorithm. Numerical studies were carried out to evaluate the performance of the new algorithm. Different FOM (figure of merit) techniques were used to evaluate the image quality. The pART algorithm shows significant image quality improvements with regard to the traditional ART in those studies.