Quantifying patterns of pathology: A characterization of the 3D spatial distribution of tau tangles and amyloid plaques in Alzheimer’s disease using multi‐modal image registration with a scattering transform

Abstract Background Volumetric measures of MRI scans have consistently shown medial temporal lobe (MTL) atrophy in individuals across the spectrum of AD. However, these measures are not considered to be specific for AD, partly because a direct link to tau and amyloid pathology has not been demonstrated (Jack et. al, 2018). To address this limitation, we developed multi‐modal image registration and analysis tools to integrate neuropathological data from 2D histology images with 3D postmortem MRI images in the same brain region, in order to quantify the density of tangles and plaques. Methods We imaged tissue excised from the MTL at 0.125 mm resolution with an 11T MR scanner (fig 1a) and manually segmented it into subregions (fig 1b). We sliced the tissue into 10um thick sections, 1mm apart, and digitally imaged them with a light microscope after staining for amyloid (6E10) or tau (PHF‐1) (fig 1c). Locations of tangles and plaques on histology were estimated with a deep convolutional neural network (UNET) trained on a smaller set of annotated data (fig 1d). To align histology images with the MRI images, we first computed scattering transforms of histology to capture high resolution textural information (fig 1e). This was then used to find both optimal spatial and intensity profile transformations between the two imaging modalities (fig 1f). We rigidly aligned MRI images to the Mai Paxinos atlas for interpretability. Results We localized tau tangles and amyloid plaques from histology sections of a person with severe AD, and in corresponding 3D MRI images of the brain tissue from the same individual. We quantified density of particles per region. Initial findings demonstrated the highest amount of tau in the entorhinal cortex, subiculum, and CA1 (fig 2) compared to other subregions in the MTL, consistent with Braak staging. No obvious differential distribution of amyloid was observed. Conclusion Our detailed localization of tangles and plaques in 3D offers the opportunity to study spatial distribution of pathology within brain regions among patients that span the spectrum of disease, which will foster more robust comparison with MRI atrophy, and potentially shed new light on the progression of AD.