Abstract Brain PET imaging has fundamentally advanced our understanding of the pathophysiology and progression of Alzheimer's disease (AD). However, the full value of this imaging modality is not realized due to issues around radiotracer dose, cost, and logistics. We believe artificial intelligence (AI) derived methods can significantly impact PET imaging in a way that makes it much more useful for molecular imaging of dementia. The goal of this project is to develop and test convolutional neural network approaches to improve the quality of brain PET imaging for two commonly used radiotracers (amyloid and tau) for dementia. We will use this to significantly reduce radiotracer dose by several orders of magnitude, freeing up PET for use in non-elderly cohorts and at repeated timepoints to better understand the early presence and subsequent progression of disease. We will also show how this new capability can enable novel paradigms for multi-tracer, single-session PET imaging. Successful completion of this study will result in validation of AI techniques for ultra-low dose PET imaging and a demonstration of how this capability can be harnessed to improve future brain molecular imaging studies to better understand the pathophysiology of AD. As such, it will provide solid, evidence- based recommendations for future clinical and research evaluation in patients with or at high risk of dementia.