Ordered-subsets EM algorithm for image segmentation with application to brain MRI

Proposes a fast expectation maximization (EM) algorithm called the ordered-subsets EM (OS-EM) algorithm for statistical model-based image segmentation. The OS-EM algorithm is based on the ordered-subsets idea which has been successfully used for image reconstruction in SPECT and PET. The OS-EM algorithm is very simple such that one needs only a few small changes in the code of the EM algorithm. The authors have applied the OS-EM algorithm to brain MRI images obtained from the MRI simulator. The results demonstrate that the OS-EM algorithm rapidly converges to a reasonable approximate solution with few iterations. Furthermore, there is no significant difference between the segmentation result with the EM algorithm and that with the OS-EM algorithm