TY - JOUR T1 - Artificial Intelligence is implemented as a tool in the daily workflow on the PET/MRI scanner to create an attenuation map for the correct PET image. JF - Journal of Nuclear Medicine JO - J Nucl Med SP - 3108 LP - 3108 VL - 61 IS - supplement 1 AU - Marianne Federspiel AU - Nadia Azizi AU - Flemming Andersen AU - Adam Hansen AU - Ian Law AU - Claes Ladefoged Y1 - 2020/05/01 UR - http://jnm.snmjournals.org/content/61/supplement_1/3108.abstract N2 - 3108Introduction: Artificial Intelligence is implemented as a tool in the daily workflow on the PET/MRI scanner to create an attenuation map for the correct PET image. Background: A  PET/MRI Siemens Biograph mMRI was installed in 2011 and used in the daily routine primarily for dementia patients. The PET/MRI is challenged by lack of a direct measure of photon attenuation correction (AC), for obtaining a correct PET image. The patients previously needed to have a low-dose CT-AC. Ladefoged et al. developed an application based on deep learning model to generate artificial CT images from MR images. The model was trained on more than 1000 patients with both MRI and CT scans. The deep learning model takes images for attenuation correction in reconstructing positron emission tomography (PET/MRI). Since May 2019 Artificial Intelligence (AI)-based MRI-AC has been implemented as a part of the daily workflow. The objective of this study was to share our experiences of the use of AI MRI-AC in clinical practice. Methods: All patients for dementia evaluation are injected with 200 MBq [18F] FDG, rest for 30 min and imaged with PET/MRI in a protocol consisting of a 10 min single-bed acquisition. The Dixon umap is centered on the brain stem and MRI sequences are focused on structural and vascular pathology. As soon as the Dixon sequence is ready it is sent to a computer node. The AI-method will generate and return a new AC map to be used during reconstruction for the correct attenuated PET image in less than 1 min. AI generated umaps are inspected before use by technologists and accuracy is followed weekly by comparing to a CT based reference. Results: Implementing AI in the daily routine for dementia patients on a PET/MRI scanner results in images that are ready to be viewed and described as soon as the patient leaves the scanner. The patients have no need for a CT scan, which saves the patients from extra radiation dose, and saves the staff some challenging time due to handling the immobile dementia patients. Conclusions: AI models, if validated for performance, offer several potential benefits to patients as well as technologists. One-stop PET/MRI for dementia patients is quick and together with the AI it is a robust and cost- effective technique for the health care system. Keywords: Artificial Intelligence Attenuation correction PET/MRI Learning Objectives: Brain Image reconstruction Positron Emission Tomography References: Andersen, F.L., Ladefoged, C.N., Beyer, T., Keller, S.H., Hansen, A.E., et al., 2014. Combined PET/MR imaging in neurology: MR-based attenuation correction implies a strong spatial bias when ignoring bone. Neuroimage 84, 206–216. Ladefoged, C.N., Hansen, A.E., Keller, S.H., Holm, S., Law, I., et al., 2014. Impact of incorrect tissue classification in Dixon-based MR-AC: fat-water tissue inversion. EJNMMI Phys. 1, 101. Ladefoged, C.N., Benoit, D., Law, I., Holm, S., Kjaer, A., et al., 2015. Region specific optimization of continuous linear attenuation coefficients based on UTE (RESOLUTE): application to PET/MR brain imaging. Phys. Med. Biol. 60, 8047–8065. ER -