Batched histogramming: A memory-efficient algorithm for PET sinogram acquisitions

When acquiring PET raw data and organizing it as sinograms, the histogramming step which converts detector pairs to PET sinograms usually consumes a large amount of memory. The memory foot print has become a serious constraint when designing a cost-efficient PET acquisition system that supports larger number of scanner rings, more time-of-flight bins, or longer scan ranges for Continues Bed Motion mode. This paper presents an algorithm that can generate PET sinograms in a memory-efficient way. Our algorithm divides the sinogram into pages. During acquisition, our algorithm converts each detector pair into a 16-bit offset value, batches them in a memory area organized by page numbers, and merges the batched values to the hard drive as needed. Imperial results show that our method effectively reduces the memory consumption for sinogram histogramming by 98%, and is fast enough to be used online.