TURBO: Automated Total-body PET Image Processing and Kinetic Modeling Toolbox
Long axial field of view (LAFOV) PET imaging requires a high level of automation and standardization, as the large number of target tissues increases the manual workload significantly. We introduce an automated analysis pipeline (TurBO, Turku total-BOdy) for preprocessing and kinetic modelling of LAFOV [15O]H2O and [18F]FDG PET data, enabling efficient and reproducible analysis of tissue perfusion and metabolism at regional and voxel-levels. The approach employs automated processing including co-registration, motion correction, automated CT segmentation for region of interest (ROI) delineation, image-derived input determination, and region-specific kinetic modelling of PET data. MethodsWe validated the analysis pipeline using Biograph Vision Quadra (Siemens Healthineers) LAFOV PET/CT scans from 21 subjects scanned with [15O]H2O and 16 subjects scanned with [18F]FDG using six segmented CT-based ROIs (cortical brain gray matter, left iliopsoas muscle, right kidney cortex and medulla, pancreas, spleen and liver) representing different levels of blood flow and glucose metabolism. ResultsModel fits showed good quality with consistent parameter estimates at both regional and voxel-levels (R{superscript 2} > 0.83 for [15O]H2O, R{superscript 2} > 0.99 for [18F]FDG). Estimates from manual and automated input functions were in concordance (R{superscript 2} > 0.74 for [15O]H2O, and R{superscript 2} > 0.78 for [18F]FDG) with minimal bias (<4% for [15O]H2O and <10% for [18F]FDG). Manually and automatically (CT-based) extracted ROI level data showed strong agreement (R{superscript 2} > 0.82 for [15O]H2O and R{superscript 2} > 0.83 for [18F]FDG), while motion correction had little impact on parameter estimates (R{superscript 2} > 0.71 for [15O]H2O and R{superscript 2} > 0.78 for [18F]FDG) compared with uncorrected data. ConclusionOur automated analysis pipeline provides reliable and reproducible parameter estimates across different regions, with an approximate processing time of 1-1.5 h per subject. This pipeline completely automates LAFOV PET analysis, reducing manual effort and enabling reproducible studies of inter-organ blood flow and metabolism, including brain-body interactions.