A Feasibility Study of Point-of-Care PET System

Objectives We studied the feasibility of the point-of-care PET (POC-PET) imaging system, designed to be compact, portable and flexible when compared to whole-body PET. This kind of scanners employs movable detectors controlled by a robotic arm so that its geometry changes during scanning. With fast image reconstruction through GPU parallel computing, it provides timely feedback to the operator who can interactively optimize the imaging task.

Methods We constructed a proof-of-principle prototype using a non-TOF PET imaging probe made of a 20×20 LYSO array (0.74×0.74×3 mm³ each) and a 4×4 MPPC array. It is mounted on a Microscribe that tracks the location and orientation. The probe is set up to be in coincidence with a half ring of Inveon PET detectors. It is randomly placed at 20 arbitrary locations around the imaging FOV to image a Na-22 source. We also simulated human-scale POC-PET scanners with TOF detectors using two different geometries in GATE. We evaluated the effect of system geometry, detector timing resolution and source-to-background ratio on the image quality by imaging a Derenzo-like phantom.

Results Reconstructed Na-22 point source images show that the tracking mechanism works well. MC simulation of the experimental setup shows near identical images. Human-scale POC-PET simulations show that sub 300 ps timing resolution is necessary in order to resolve lesions with contrast ratio as low as 6:1 and to avoid significant artifact in images that are acquired from limited angles. Our GPU based fast image reconstruction handles 0.35 million list-mode events per second in one iteration.

Conclusions We proposed a new class of PET systems that is aiming at POC applications. A prototype was built to demonstrate some technologies that are essential for POC-PET systems. MC simulation shows resolution of 4 mm hot lesions under practical imaging conditions. Our image reconstruction handles dynamic geometry well and can potentially provide near-real time imaging capability.

Research Support This work is supported in part by the NIH (CA136554), NSF (DBI-1040498), and Mallinckrodt Institute of Radiology internal funds. The WU Center for High Performance Computing are funded in part by NIH (RR031625, RR022984)