Comparative assessment of energy-based methods of compensating for scatter and lead X-rays in Ga-67 SPECT imaging

Two energy-based methods of compensating for scatter, high-energy contamination, and lead X-rays in Ga-67 SPECT images were optimized and compared on the basis of their performance in least-squares estimation of tumor activity uptake. The artificial neural network (ANN) and linear, general spectral (GS) correction methods developed in our group both make use of images stored in many energy windows to estimate the primary, unscattered counts attributable to each of the principal photopeaks of Ga-67 at 93, 185, and 300-keV. Projection images of an anthropomorphic phantom containing background organ uptake and spherical 'tumors' of different size and activity in 14 possible locations typical of 72-hour Ga-67 lymphoma studies were simulated in 55 energy windows spaced irregularly in the range, 60-370 keV. Seven of the sphere data sets were used for optimizing the correction methods, while 12 Poisson noise realizations of each of the other seven sphere data sets were used for performance evaluation. The ANN method was trained to minimize the mean-squared error of corrected pixel values, while the GS method was optimized specifically for the tumor activity estimation task.