Abstract
Partial volume effect (PVE) is one of the major degradation effects for quantitation in PET due to finite system resolution and spatial sampling. To avoid quantiation bias casued by PVE in PET, correction for PVE is necessary but not routinely applied in clinical practice. There have been many different correction schemes proposed in the literature, including methods of deconvolution, and geometric transfer matrix (GTM). The popular and effective PVC method of geometric transfer matrix (GTM) was proposed by Rousset in 1998, and anatomic information (ex. MRI, CT) is needed in modeling PVE from different regions. Due to the requirement for anatomical images, segmentation and registration between PET and anatomical images are needed, and thus mis-segmentation and mis-registration would potentially introduce errors in the PVC results. In this thesis, a modified GTM method (mGTM) based on previous GTM algorithm was proposed, and the performance of the proposed mGTM was compared to other popular PVC methods from literatures. The major difference between mGTM and GTM is from region-based correction (GTM) to voxel-based processing (mGTM). Therefore, no anatomical information is required. Nevertheless, due to huge amount of data, the computation load for mGTM method is heavy. Thus the computational process in solving the original linear equations is simplfied by taking only a small set of variables each time. The results demonstrated that the proposed mGTM generated more accurate quantitation and better image recovery than other PVC methods and without any need for anatomical information. Future work will include more simulation and clinical data to verify the performance of mGTM method in 3D.