Alzheimer’s Disease (AD) is the most common cause of cognitive decline in aging, with a sharp rise in prevalence from 60 years, affecting about 30% of the population above 90 [1]. The big challenge is that degeneration in the brain is ongoing for years before clinical symptoms. Hence, early detection of AD, prior to extensive irreversible brain damage, is critical for any treatment to be effective. Further, early detection is necessary for (a) selection of patients to randomized control trials (RCT), (b) optimizing treatment and care, (c) prognosis and information for patients and caregivers, and (d) initiation of preventive measures. Tools are also required to monitor treatment effects. Hence, methods to detect early brain changes in AD are necessary in research and the clinic. Multiple initiatives focus on this important question, and we and others have shown that quantitative MRI to some degree can predict conversion from mild cognitive impairment (MCI) to AD, especially by tracking changes over time [2, 3]. However, a key challenge is that sensitivity and specificity are too low and drops when based on a single examination. A second key challenge is that quantitative MRI is rarely included in clinics. Common praxis in neuroradiology is the use of visual rating scales, which does not take advantage of the vast amount of information embedded in an MR image. We propose to address these key challenges: (1) Use machine learning to develop novel quantitative MRI methods to detect early AD with higher sensitivity and accuracy than state-of-the art methods. (2) Implement and evaluate a system for efficient use of these methods in a clinical setting at Oslo University Hospital (OUS). Machine learning has shown remarkable performance on MRI-based AD classification (Wen et al., 2020). This has not translated to successful clinical use (Nagendran et al., 2020), due to obstacles such as lack of large representative training data, computational constraints and lack of integration with hospital systems. PredictAD meets these challenges. One unique aspect is that a novel machine learning approach, including several independent validation steps, is used on large international samples of patients and controls (n = 48.500), coupled with unique and realistic Norwegian biomarker-informed clinical data (n = 1100). Critically, development and testing of the machine learning algorithms are tightly integrated with clinical application and validation. We will run an experiment involving clinicians at OUS and Ahus, to assess how diagnosis and prognosis can be improved in a clinical setting. The final output will be clear recommendations for best practice regarding use of quantitative MRI in AD research and clinic. In addition, the project aims to strengthen national and international collaboration and networks for AD research, the connection between research and clinic in Norway, and increase national expertise in advanced MR imaging of AD. We will also work directly with the researchers behind the most frequently used software suite for quantitative MRI, FreeSurfer, developed at Harvard Medical School, in the generation of the new tools. This will secure very valuable input to the project, and greatly increase the chances for uptake of the project output in the international research community. All computer code, procedures and results generated in the project will be freely available to the community. We believe the tight integration throughout the project of user perspectives, clinicians and international methods developers will maximize the likelihood that the project will have real impact on users, clinics and academia.