Real-time event stream correction for patient motion in clinical 3-D PET

Stationary phantoms are typically used to demonstrate intrinsic image resolution in PET. Yet, real image resolution in most clinical PET studies degrades by a reported factor of 3 to 5 due to unintentional patient motion. Motion tracking instruments are available (Polaris from NDI) to measure complex 3-D and 6-D movements in real time. Still, clinical sites demand high patient throughput. Time available for post-acquisition data processing is limited. This article presents an architectural concept for on-line PET data acquisition systems that deliver the benefits of motion correction consistent with high patient throughput. Such systems provide real-time processing for coincidence-by-coincidence correction of patient motions. For this concept, five processing stages are described. Each stage feeds the next with 64-bit (or larger) coincidence event packets and operates continuously at up to 10M events/sec or faster. Stage 1 incorporates an event correction factor (ECF) into each incoming detector-pair packet. For example, each 16-bit block-floating ECF corrects for normalization, dead time, etc. and is used later in Stage 5. Stage 2 converts detector-pair content of the packets into xyz-pair content-effectively changing each scintillator element index into a stationary 3-D coordinate index. Each x, y and z value is a 12-bit signed integer. Stage 3 transforms xyz pairs into x'y'z' pairs by application of motion tracking data. Each x'y'z' 3-D coordinate system remains fixed relative to the indicated portion of the patient body-e.g. chest, pelvis, head, etc. Stage 4 converts x'y'z'-pair packets into bin-address packets. The 37-bit bin address references unique elements (32-bit bins) in the projection array. Stage 5 performs weighted histogramming in which the ECF value-carried along with each packet from Stage 1-is added to (subtracted from) the indicated bin value for each prompt (delayed) event.