Quantitative accuracy of SPECT reconstructions with a voxelized CT-based partial volume effect correction

For many clinical applications, it is important to know not only the total activity in an organ of interest, but also the details regarding this activity distribution. In this context, the term “quantitation” would refer to the recovery of not only the total activity but also its voxelized concentration. In this paper, we analyze the quantitative accuracy of activity distributions in ^99mTc SPECT/CT images using data from physical phantom experiments. Our acquisitions followed cardiac (^99mTc-MIBI) and oncology (^99mTc-MDP) clinical acquisition protocols and were aiming at providing the medical community with information about absolute ^99mTc activity distribution in the myocardium (for the diagnosis of a balanced triple vessel disease) and in tumors (for evaluation of their aggressiveness). Our algorithm includes the following steps: (i) Reconstruction of initial images with corrections for attenuation, scattering, and resolution loss; (ii) Creation of a template (numeric phantom) covering not only the investigated region of interest (ROI), but also other active areas. allowing us to simultaneously take into account both spill-in and spill-out effects; (iii) Projection and reconstruction of this template (with the same corrections as for the initial images); (iv) Voxel-by-voxel partial volume effect correction (PVC). After applying all the aforementioned corrections, we recovered total activities inside both heart and tumor models with errors less than ∼6% for 33ml containers and ∼9% for 120ml heart insert. PVC substantially improved not only the accuracy of the recovered total activity (up to 40%), but also its distribution inside the ROI. The current version of this PVC technique makes an assumption that the ratios of activity between different compartments are known. Although this assumption may hold in some clinical situations, for example for tumors in areas with low background activity, a- - t the next stage we plan to test the use of the initial image reconstruction for creation of the template.