Assessment of the effect of 3D resolution recovery during SPECT image reconstruction on quantification of small liver tumors: A simulation study

摘要

Non ideal collimation and detector blurring are the main causes of resolution loss in single photon computed tomography (SPECT) that affect both qualitative and quantitative interpretation of SPECT images. In this research we evaluate the importance of resolution recovery in determining the absolute amount of activity in small liver tumors by comparing three different approaches: ordered subsets expectation maximization (OSEM), OSEM with 2D resolution recovery and OSEM with 3D resolution recovery. Two different sets of projections were created. Model based projections were created by layer by layer blurring of activity phantom using known distance dependent Gaussian functions. Simple ray sum of this blurred phantom was driven to create projections. This simulates noise free projections with completely known system response. The other set of projections was created using SIMIND Monte Carlo simulator. Eight small tumors (1×1×1 cm) were considered on different parts of NCAT liver phantom in such a way that all appear in same plane in a sagittal slice. Vertical distance between each couple of tumors was varied between 1 and 4 pixels (equal to 3.3 and 13.3 mm). To evaluate the value of resolution recovery during reconstruction, OSEM with the ability of modeling both 2D and 3D system response model during reconstruction was developed and used. Both 2D and 3D resolution recovery results in better quantitative results compared to simple OSEM. 3D resolution recovery, however reconstruct images with much higher quantitative accuracy. Only in 3D resolution recovery it is possible to separate all couple of tumors in axial direction. The amount of activity recovery depends on the number of iteration. Using 8 iterations and 8 subsets in each iteration, 3D resolution recovery results in 8% better activity recovery compared to 2D and 13% compared to simple OSEM. This amount is 27% and 46% for 25 iterations.