1006 Introduction: Prostate-specific membrane antigen (PSMA)-ligand PET imaging has shown to be valuable for imaging men with prostate cancer. The problem of inter-observer variability related to subjective visual analysis and the lack of objective quantitative methods make it difficult to compare results from different centers. There is therefore a need for automated methods to quantify PET/CT findings. The aim of this study was to develop an artificial intelligence (AI) tool for detection and quantification of primary prostate tumors, bone metastases, and lymph node lesions in PSMA PET/CT studies, and to evaluate the effect of the AI tool on inter-observer variability. Methods: The test group consisted of 10 patients referred for PSMA PET/CT for initial staging at Skane University Hospital, Lund and Malmo, Sweden during 2019-2020. Patients were administered with 4 MBq/kg [18F]-PSMA-1007 and after 2 h scanned on a Discovery MI PET/CT (GE healthcare, Milwaukee, WI).Three nuclear medicine specialists independently segmented primary prostate tumors, bone metastases, and lymph node lesions in the 10 studies, first without AI support and then, more than eight weeks later, with segmentations proposed by the AI tool. The new AI tool relies heavily on the CT-based organ segmentation tool from work by Tragardh et al. (EJNMMI Phys. 2020;7:51). To reduce the problem of PET/CT misalignment, each pixel is assigned to a close local maximum using a flooding algorithm. Uptake in a pixel is assumed to originate from the organ of the corresponding local maximum.Prostate tumors: The lesion threshold is set to standard uptake value (SUV) 2 + first quartile of the prostate uptake, but at most SUV 5. Any high uptake region connected to the prostate is included unless it originates from bone, bladder or gastrointestinal tract.Bone metastases: For each of the following groups, vertebra, rib, scapula, radius, fibula, patella, hand, humerus, clavicle, tibia, foot, hip, femur, ulna and sternum, the lesion threshold is set to SUV 3 * median bone uptake, restrained between SUV 2 and SUV 10. Lymph node lesions: The model from Borrelli et al. (Clin Physiol Funct Imaging. 2021;41:62) was used to directly segment lymph node lesions. Finally, lesions with total lesion uptake (TLU) less than 2.0 were removed. The McNemar test was used to analyze the difference in patients with increased and decreased variability between observers without and with support of AI. Results: To measure the inter-observer variability in TLU measurements for each patient, the relative difference between pairs of observers was computed. For both prostate TLU and bone TLU, the average of the three relative differences, decreased for all 10 patients with the support of AI (p=0.004). For lymph node lesions, the average relative differences decreased for 7 patients, increased for 2 patients and was unchanged in 1 patient with the support of AI (p=0.18). Conclusions: Physicians supported by an AI tool for automated quantification of PSMA-PET/CT studies showed significantly less inter-observer variability in the quantification of primary prostate tumors and bone metastases than when performing a completely manual analysis. This AI tool may help in facilitating comparison of studies from different centers, pooling data within multicenter trials and performing meta-analysis. The AI tool developed in this project is available upon reasonable request for research purposes at www.recomia.org.