A real-time TOFPET slice-backproject engine employing dual AM29116 microprocessors

Tomographic backprojection for both conventional and TOF PET systems is data intensive, and without special hardware can require an unacceptable amount of time. We have designed, constructed, and are using a prototype slice-backproject engine that has improved our throughput by a factor of 30. The design is centered about two microprogramable 16-bit processors and a 16 X 16 combinational multiplier with an instruction rate of 150 to 200 nsec. This paper describes the engine architecture as it relates to the backproject algorithm, as well as design features that enable its use as a general processor. The use of two processing elements permits shared and concurrent operations to be performed at the microcode level. Finally, the engine is reviewed in the context of a system architecture employing multiple engines to backproject several slices in parallel