Improving the sensitivity and specificity of MRI-based biomarkers in Alzheimer's disease

Finding a cure for Alzheimer's disease (AD) is one the greatest scientific challenges of our time. There is a growing realization that experimental treatments should target the earliest, presymptomatic stages of disease. But the cost of conducting clinical trials in participants who may not develop symptoms of AD for years can be prohibitive. There is a growing need for more effective biomarkers of AD, particularly biomarkers of therapeutic efficacy that can detect slowing or reversal of AD-related changes due to treatment as early in the clinical trial as possible. These biomarkers must be as sensitive as possible to disease progression and also account for AD heterogeneity, i.e., the fact that the majority of individuals who have AD pathology also have one or more concomitant pathologies that may affect their ability to respond to experimental treatments for AD. This proposal focuses on deriving effective presymptomatic and early symptomatic AD biomarkers from magnetic resonance imaging (MRI), the imaging modality that provides the most direct evidence of neuritic and neuronal loss in neurodegenerative disease. Rather than propose new MRI acquisition protocols, we focus on the most commonly collected type of MRI scan in AD research (T1-weighted 3D gradient echo scans with approximately 1x1x1mm3 resolution) and use advanced computational analysis to quantify changes in the subregions of the medial temporal lobe (MTL), the brain region that associated with early stages of AD pathology as well as with early stages of multiple concomitant non-AD pathologies. Aim 1 will develop and validate advanced algorithms that combine conventional multi-atlas segmentation with deep learning to reliably extract small subregions of the MTL, such as Brodmann area 35, explicitly accounting for anatomical variability in the MTL. Aim 2 will correlate quantitative digital pathology measures derived at autopsy with antemortem MRI to discover distinct patterns of change that we hypothesize are associated with concomitant pathologies in AD, including TDP-43 pathology, alpha-synucleinopathy, non-AD tauopathies, and cerebrovascular disease. Aim 3 will apply deep learning to improve the sensitivity of measures of change in longitudinal MRI, hypothetically leading to a more sensitive early marker of treatment effectiveness in trials targeting preclinical AD than existing cognitive and imaging-based measures. Taken together, improved precision of segmentation (Aim 1), determination of spatial patterns of AD and concomitant non-AD pathology (Aim 2), and advanced longitudinal measurement methodology (Aim 3) will optimize MTL subregional sensitivity and specificity for the earliest neurodegenerative changes of AD, providing a potentially critical therapeutic efficacy measure to accelerate clinical trials of disease modifying treatments in early AD.