Acceleration of MAP-EM Algorithm via Over-Relaxation

In emission tomography, such as positron emission tomography (PET) and single-photon emission computed tomography (SPECT), statistical reconstructions outperform FBP (filtered back-projection) [1], because of their ability to model noise, system geometry and imaging physics, in addition to incorporating prior information that is related to the object. One of the most widely used statistical reconstruction methods is the ML-EM (maximum likelihood expectation-maximization) algorithm [2]. The ML-EM algorithm is hampered by its slow convergence rate and has difficulty with the ill-condition problem [3].
Source: Computerized Medical Imaging and Graphics - Category: Radiology Authors: Source Type: research