Joint Estimation for Incorporating MRI Anatomic Images into SPECT Reconstruction

To improve SPECT reconstruction using spatiallycorrelated magnetic resonance(MR) images as a source of side information, one must account for mismatch between MRI anatomical information and SPECT functional information. We investigate an approach which incorporates the anatomical information into SPECT reconstruction by using region labels representing the anatomical regions extracted from MRI. Each SPECT pixel corresponds to one region label. Both SPECT pixel mean intensities and region labels are jointly estimated by a penalized MaximumLikelihood criterion using an iterative Space-Alternating Generalized EM algorithm. The likelihood function incorporates both the SPECT noise distribution and the MRI side information measurement statistics. Since the region labels are estimated jointly from both segmented MRI and SPECT projection data, only those anatomical regions that match SPECT functional regions are represented by the estimated labels, and are used to constrain the SPECT reconstruction. The artifacts due to the mismatched MR anatomical region information are reduced using joint estimation. By comparing image quality and the Bias us. Variance tradeoffs, we see that the joint estimation has the potential to improve the SPECT estimation result.