Diagnosing the undiagnosable: studies of Alzheimer disease mimics and confounders via "neuropathometry" of dissection photos with 3D scanning - Project Summary Title: Diagnosing the undiagnosable: studies of Alzheimer disease mimics and confounders via neuropathometry of dissection photos with 3D scanning Summary: While most patients with late life dementia have Alzheimer’s disease (AD), there are conditions that overlap or even mimic AD, confounding clinical diagnosis, and thus representing a barrier to accurate predictions of rate of progression and to effective therapeutics. Examples of these diseases include concomitant TDP-43 pathology, Dementia with Lewy bodies (DLB), and microvascular lesions associated with poorly defined white matter lesions. A critical barrier to studying these diseases is that there currently is no reliable premortem biomarker. Here we propose a collaboration with two Alzheimer’s Research Centers to evaluate anatomical signatures of these three conditions, in contrast to AD, in order to enable research into them, and ultimately port back to MRI in order to directly enhance clinical care. Specifically, we propose to use advanced machine learning (ML) techniques to perform volumetric photographic scanning post mortem (at autopsy), on patients seen at the Massachusetts Alzheimer Disease Research Center (MADRC). Reconstructing imaging volumes from dissection photographs, which are routinely acquired at brain banks and neuropathology departments, will enable us to correlate neuropathology with macroscopic measurements (e.g., volume and shape of brain structures, cortical thickness) without the need for magnetic resonance imaging (MRI) data. This is crucial because diagnostic MRI is not always acquired close to autopsy, or at all, and ex vivo MRI is expensive, technically challenging, and not available at many research sites. Therefore, our technique has the potential of greatly increasing sample sizes, especially with asymptomatic individuals who were not scanned in life, and who would likely manifest the earliest and purest neuropathological changes. Our tools will combine ML with 3D shape scanning, which is an increasingly inexpensive technology ($1,000 - $10,000 for a scanner), to produce very accurate reconstructions of the brain shape. Moreover, we will also build an “atlas” version of the tool, that replaces 3D scanning by a probabilistic atlas, thus enabling analysis of retrospective data. We will develop the tools in collaboration with a second ADRC, the University of Washington ADRC, which has slice photographs for approximately one thousand cases. The new tools will be used to closely study a prospective cohort at MADRC, consisting of 200 subjects. We seek to identify neuroimaging signatures of the AD mimics mentioned above, which can be ported to in vivo MRI scanning. Moreover, we will also distribute and maintain the tools as part of our neuroimaging package FreeSurfer (over 40,000 worldwide licenses), so they can be used by research sites around the world to augment neuropathology with macroscopic morphometric measures at little or no cost.
