Radionuclide scatter estimation using EGS4-generated convolution kernels

In nuclear medicine, the limited energy resolution of scintillation detectors hinders identification of scatter photons, leading to inaccurate quantitation in SPECT imaging. The authors objectives were to develop a method for estimating the detected scatter counts in an arbitrary energy window and to incorporate that method into iterative SPECT reconstruction. The scatter estimation method calculates the expected scatter projection data by forward projecting an “effective” scatter source distribution through an attenuation distribution. This “effective” scatter source distribution is approximated by convolving an estimated primary source distribution with a kernel that is pre-calculated using the EGS4 Monte Carlo code. This scatter estimation method was confirmed by independent, Monte Carlo-simulated acquisitions of several phantom configurations, including point sources, uniform tanks, and spheres within nonuniform tanks. In addition, this scatter estimation method was incorporated into an iterative Maximum Likelihood, Expectation Maximization (MLEM) reconstruction code and used to reconstruct scatter-corrupted data to demonstrate the effects of scatter correction. Line profiles through Monte Carlo-simulated data and scatter data estimated using this convolution-projection technique demonstrated good agreement for ^99mTc. Uniform tanks reconstructed with this scatter estimation scheme were within 2% of known activity concentrations. Without scatter correction, reconstructed activity concentrations were overestimated by more than 20%. In conclusion, the authors have developed a method for estimating detected scatter counts that is suitable for iterative SPECT reconstruction