Comparing geometries for a PET system with 3-D photon positioning capability

We compared two PET system geometries, a box and a cylinder, for a novel PET 3-D detector design with axial, transaxial, and depth of interaction resolution of 1 mm, 1 mm, and 3 mm, respectively. Monte Carlo simulations have shown that arranging this detector into an 8×8×8 cm³ FOV box has twice the sensitivity of a conventional cylinder of comparable dimensions (8 cm diameter, 8 cm axial length) due to a reduction in inter-module gaps. We compared image reconstruction performance for the box and the cylinder by computing the mean square error (MSE) for a weighted least squares estimator and contrast recovery coefficient (CRC) for rod phantoms with rods located at the radial center and radial offset of 2 cm. The contrast was comparable (near 100% CRC) for both box and cylinder. However, the cylinder showed slightly better contrast uniformity (CRC varied 2% for cylinder, 8% for box) as the radial position of the lesion was varied. The box had a MSE that was 7-14 times better than the cylinder system. We performed an ROC study by computing the detection SNR for computer observer models. A non-prewhitening and a prewhitening test statistic were computed to provide an upper and lower bound for the performance of human observers. The prewhitening filter detection SNR for the cylinder design was superior (by <15%) to the box at the center position. For all other source locations, the box showed an improvement (by >20%) in detection SNR. Finally, we reconstructed a resolution phantom with 3-D data for both systems and showed that both systems were qualitatively comparable.