Multimodal magnetic resonance imaging predicts regional amyloid-β burden in the brain

Alzheimers disease (AD) is the most common cause of dementia and identifying early markers of this disease is important for prevention and treatment strategies. Amyloid - {beta} protein deposition is one of the earliest detectable pathological changes in AD. But in-vivo detection of amyloid - {beta} using positron emission tomography (PET) is hampered by high cost and limited geographical accessibility. These factors can become limiting when PET is used to screen large numbers of subjects into prevention trials when only a minority are expected to be amyloid- {beta} - positive. Structural MRI is advantageous; as it is relatively inexpensive and more accessible. Thus it could be widely used in large studies, even when frequent or repetitive imaging is necessary. We used a machine learning, pattern recognition, approach using intensity-based features from individual and combination of MR modalities (T1 weighted, T2 weighted, T2 fluid attenuated inversion recovery [FLAIR], susceptibility weighted imaging) to predict voxel-level amyloid- {beta} in the brain. The MR- amyloid {beta} relation was learned within each subject and generalized across subjects using subject-specific features (demographic, clinical, and summary MR features). When compared to other modalities, combination of T1-weighted, T2-weighted FLAIR, and SWI performed best in predicting the amyloid- {beta} status as positive or negative. T2- weighted performed the best in predicting change in amyloid- {beta} over two timepoints. Overall, our results show feasibility of amyloid- {beta} prediction by MRI.