Abstract We previously showed that the injected activity could be reduced to 1 MBq/kg without significantly degrading image quality for the exploration of neurocognitive disorders in 18F-FDG-PET/MRI. We now hypothesized that injected activity could be reduced ten-fold. We simulated a 18F-FDG-PET/MRI ultra-low-dose protocol (0.2 MBq/Kg, PET ULD ) and compared it to our reference protocol (2 MBq/Kg, PET STD ) in 50 patients with cognitive impairment. We tested the reproducibility between PET ULD and PET STD using SUVratios measurements. We also assessed the impact of PET ULD for between-group comparisons and for visual analysis performed by three physicians. The intra-operator agreement between visual assessment of PET STD and PET ULD in patients with severe anomalies was substantial to almost perfect (kappa > 0.79). For patients with normal metabolism or moderate hypometabolism however, it was only moderate to substantial (kappa > 0.53). SUV ratios were strongly reproducible (SUVratio difference ± SD = 0.09 ± 0.08). Between-group comparisons yielded very similar results using either PET ULD or PET STD . 18F-FDG activity may be reduced to 0.2 MBq/Kg without compromising quantitative measurements. The visual interpretation was reproducible between ultra-low-dose and standard protocol for patients with severe hypometabolism, but less so for those with moderate hypometabolism. These results suggest that a low-dose protocol (1 MBq/Kg) should be preferred in the context of neurodegenerative disease diagnosis.