A direct time-of-flight reconstruction for whole-body single-bed RPC-PET: Results from lesion and anthropomorphic simulated data

A single-bed, whole-body positron emission tomograph based on resistive plate chamber detectors has been proposed (RPC-PET). It has been shown by simulation that RPC-PET with an axial field-of-view (AFOV) of 2.4m is feasible and yields an absolute sensitivity enhancement of c. one order of magnitude superior to that of typical cylindrical, crystal-based PET scanners, with 16cm AFOV. In addition to its time-of-flight (TOF) advantage, RPC-PET offers potential very-high spatial resolution at the detector level. This must be properly handled by a fully-3D reconstruction algorithm capable of processing the very inclined lines-of-response (LOR) from large AFOV systems such as RPC-PET. For RPC-PET, these are acquired within a single-bed examination. Consequently, one or several segmented images representing the full human body during reconstruction are necessary. In this paper, we show that a direct-TOF implementation of the ordered subset expectation maximization (OSEM) algorithm allows for all events acquired with RPC-PET to be directly processed without rebinning and directly inserted inside the object image by means of a TOF-kernel. Such kernel (1) avoids typical (and slow) voxel-wise image navigation, (2) respects the time uncertainty dictated by the imaging RPC detectors, and (3) permits handling list-mode data iteratively with CT-based attenuation correction and OSEM computed both on-the-fly. We present reconstructed result from blind simulations of a single-bed, whole body PET system, corresponding to (1) six simulated spherical sources immersed in a homogeneous activity background, (2) an anthropomorphic phantom including a realistic grey-to-white matter uptake ratio of four, (3) and an additional anthropomorphic phantom with oncological lesions. We show that overall specificity for lesion detection may be increased (false positives decreased) by multiple similar, but independent reconstructions that eliminate false candidates arising from image statistical noise.