Individual functional brain mapping for biomarker discovery in Alzheimer's

PROJECT SUMMARY: Alzheimer’s disease (AD) affects over 6 million Americans, and its incidence is projected to double by 2050 due to an aging population. In the fight against AD, there is a pressing need for novel biomarkers to 1) identify clinical trial participants at risk of decline and 2) identify and track patients eligible for emerging treatments. Gold standard AD biomarkers require positron emission tomography (PET) imaging or cerebral spinal fluid (CSF) collected via lumbar puncture. These procedures are expensive and/or invasive, presenting a barrier to widespread adoption. Blood-based biomarkers are under development but are not yet validated and may benefit from combination with other biomarkers. A principal goal of the Alzheimer’s Disease Neuroimaging Initiative (ADNI) 3 is to promote development of novel AD biomarkers, including the use of functional magnetic resonance imaging (fMRI). FMRI is used to study the functional connectivity (FC) and organization of the brain, and fMRI studies have revealed functional brain changes associated with AD. FMRI- based biomarkers of AD could complement existing biomarkers by providing a non-invasive, first-line screening method before PET and CSF are collected. FMRI could also be combined with blood biomarkers and established structural MRI markers—all of which can be routinely collected clinically—to construct powerful and widely available multimodal biomarkers. Despite all of this, no fMRI-based biomarker for AD exists to date. This is in part due to the high noise levels of fMRI and the common use of naive statistical methods, which together lead to noisy estimates of FC and other functional brain features. Two conventional workarounds—averaging many subjects or collecting hours of data on individual subjects—are not viable clinically. This project will address this gap by developing computationally efficient Bayesian techniques with high accuracy and deploying those methods for fMRI-based biomarker discovery in AD. Our models leverage information across subjects via population-derived priors or “templates”, which are previously estimated, to extract nuanced and precise functional brain features in individuals. These models avoid the need for burdensome prolonged scans. They can be fit to data from a single subject at a time, making them clinically viable and computationally advantageous. To maximize the benefits of hierarchical modeling, we utilize grayordinates data, a recent technological advance in image processing that improves inter-subject anatomical alignment. To deploy these techniques effectively in multi-site datasets like the ADNI, image harmonization is necessary to avoid confounding site effects. Existing harmonization methods can be applied to fMRI summary measures, but are not applicable to fMRI time series, which are a complex mixture of latent features. To address this critical gap, we will develop a novel harmonization method for fMRI time series data, with high potential impact on fMRI processing pipelines. Finally, we will analyze fMRI data from ADNI to extract functional brain features and build novel fMRI-based AD biomarkers. While our focus is AD, this project will have broad implications for fMRI-based research and care.