RT Journal Article SR Electronic T1 Ultra-Fast Reconstruction of Short List-Mode PET Data Frames for Real-Time Visualization and Processing JF Journal of Nuclear Medicine JO J Nucl Med FD Society of Nuclear Medicine SP 570 OP 570 VO 61 IS supplement 1 A1 Spangler-Bickell, Matthew A1 Deller, Timothy A1 Jansen, Floris YR 2020 UL http://jnm.snmjournals.org/content/61/supplement_1/570.abstract AB 570Introduction: Clinical PET reconstruction typically begins after the entire frame of data has been acquired. In this paper we present a framework for performing reconstructions of very short frames of list-mode data (∼1 sec) faster than real-time (i.e. < 1 sec), with the objective of using this framework during the scan to enable various applications:• The technologist can observe the tracer distribution in near real-time during the scan and verify the patient positioning, the injection success, any motion, etc., and intervene accordingly.• Perform data-driven motion estimation so that as soon as the scan is completed a fully motion corrected reconstruction can be performed.• Alert the technologist if motion exceeds a threshold.• Conduct a whole body survey over a few seconds to quickly see the tracer distribution, identify areas of interest, and adjust the subsequent bed durations accordingly.• For gated cardiac imaging, quick reconstructions of the gates can aid in determining which gate most closely matches the attenuation map.The developed framework runs on top of the PET research toolbox distributed by GE Healthcare (Chicago, IL). Method: A fast TOF-based ray-tracing projector is used within a list-mode (LM) reconstruction algorithm. Batches of LM data corresponding to a chosen frame duration are reconstructed using MLEM without subsets. 2-5 iterations are performed. Attenuation correction is applied, unless motion estimation is to be performed as it might bias the motion estimates. Randoms correction is calculated independently for each detected event, based on the per-second single events information. Scatter correction is not performed. For visualisation purposes, a 1 sec duration has been found to be adequate. To estimate head motion, frame durations as low as 0.3 sec provide sufficiently accurate motion estimates [1]. Clinical Results: A standard clinical 18F-FDG brain data set was acquired on a Discovery MI PET/CT (GE Healthcare). Frames of 1 sec duration were reconstructed, each taking approximately 0.2 sec on an 8-core CPU desktop system. In figure 1, the patient’s hand can be seen entering the field-of-view, which was not observed during the scan.In figure 2, a patient undergoing a brain scan on a SIGNA PET/MR (GE Healthcare) spoke to the technologist, resulting in head motion. Motion was estimated from the frames, and a fully motion corrected LM reconstruction was then performed (figure 3).Using data from a whole body 18F-FDG scan, very short frames of 0.3 sec each were reconstructed from several bed positions over the abdomen and thorax and stitched together (figure 4). This demonstrates the feasibility of an ultra-fast whole body survey. Conclusions: A framework for performing ultra-fast reconstructions (< 1 sec) of very short frames (~ 1 sec) has been presented, allowing near real-time reconstruction of the tracer distribution during a PET scan. Such a framework opens many avenues, including real-time feedback to the technologist during the scan, data-driven motion correction in the head and abdomen, and rapid whole body surveys. Future work includes extracting non-rigid deformation parameters for data-driven motion correction with respiratory and cardiac motion. Fig. 1: Example reconstructions of 1 second frames. The patient’s hand can be seen entering the FOV. Fig. 2: Selected frames of 1 sec duration where the patient can be seen speaking during the scan. Fig. 3: Left: Plots of the motion estimates. The patient spoke at about 380 sec. Right: List-mode reconstructions of the full data set without and with motion correction (MC). Fig. 4: Very short frames of 0.3 sec each from several bed positions to demonstrate a whole body survey. Acknowledgement: We gratefully acknowledge patient data acquired at KU Leuven. References: [1] M. G. Spangler-Bickell, et al, “Effect of image noise on registration in PET brain imaging,” in IEEE MIC Proceeding, 2019