Randoms simulation for dual head coincidence imaging of cylindrically symmetric source distributions

High detector singles rates in Dual Head Coincidence Imaging (DHCI) produce significant randoms fractions in the acquired data. Although some DHCI systems provide a means to estimate randoms, randoms correction is not routinely performed in DHCI. We simulate random coincidence data from cylindrically symmetric objects in a DHCI system with 51-cm×37-cm detectors of 9.5-mm thick NaI(TI) spaced 63 cm apart. The randoms are normalized as if they were true events and reconstructed using 3D reprojection. The reconstructed normalized random coincidences are broadly distributed throughout the field of view (FOV). Axially, they peak toward the center of activity; they have a transaxial structure that is weakly object-dependent. Transaxial randoms image profiles of cylindrically symmetric objects have an offset at the transaxial FOV edge and a broad peak toward the FOV center. Centered spheres and cylinders can produce cold lines along the axis of rotation, whereas flood sources create randoms distributions that peak along the axis of rotation. A Monte Carlo investigation of the bias and variance that randoms contribute to images of a 20-cm cylinder acquired for 20 minutes at a singles rate of 400 photopeak kcps per detector is described. Without attenuation correction, randoms contribute an average bias equal to 9.2% of the mean true+scatter image intensities over the cylinder. Attenuation correction without randoms subtraction amplifies this error. Randoms subtraction with attenuation correction does not significantly boost the image noise for the 20-cm cylinder