Deep Learning for Alzheimer’s Disease: Mapping Large-scale Histological Tau Protein for Neuroimaging Biomarker Validation

Abstract

Deposits of abnormal tau protein inclusions in the brain are one of the pathological hallmarks of Alzheimer's disease (AD) and are the best predictor of neuronal loss and clinical decline. As such, imaging-based biomarkers to detect tau deposits in-vivo could leverage AD diagnosis and monitoring from earlier disease stages. Although several PET-based tau tracers are available for research studies, validation of such tracers against direct detection of tau deposits in brain tissue remain unresolved. Large-scale voxel-tovoxel correlation has been challenging because of the size of the human brain, deformation caused by tissue processing that precludes registration and the amount of tau inclusion. In this study, we created a semi-automated computer vision pipeline for segmenting tau inclusions in billion-pixel digital pathology images of whole human brains, aiming at generating quantitative, tridimensional tau density maps that can be used deciphering the distribution of tau inclusions along AD progression and validate PET tau tracers. Our pipeline comprises several pre and post-processing steps developed to handle the high complexity of these brain digital pathology images. SlideNet, a convolutional neural network designed to work with our large-datasets to locate and segment tau inclusions is the heart of the pipeline. Using our novel method, we have successfully processed >500 slides from two whole human brains, immunostained for two phospho-tau antibodies (AT100 and AT8) spanning several Gigabytes of images. Our network obtained strong tau inclusion segmentation results with ROC AUC of 0.89 and 0.85 for AT100 and AT8, respectively. Introspection studies further assessed the ability of our trained model to lean tau-related features. As the final results, our pipeline successfully created 3d tau density maps that were co-registered to the histology 3d maps.

3D Mapping of Neurofibrillary Tangle Burden in the Human Medial Temporal Lobe

“…To our knowledge, this work is the first to perform 3D mapping of tau NFT pathology in a sizable collection of human brain specimens ( n = 18). In a recent preprint, Alegro et al [4] describe a pioneering pipeline for 3D tau density maps generation in whole brains and present results for two brain specimens. They use deep learning to extract and quantify tau inclusions from anti-tau immunohistochemistry slides and use registration to ex vivo MRI to reconstruct 3D maps.…”

“…They use deep learning to extract and quantify tau inclusions from anti-tau immunohistochemistry slides and use registration to ex vivo MRI to reconstruct 3D maps. Alegro et al [4] perform ex vivo MRI and CT in situ , an advantage over the current work since this approach eliminates MRI artifacts due to air bubbles and makes co-registration with in vivo MRI simpler. However, in situ scanning is not always feasible to integrate into existing autopsy protocols, including at the brain banks used in our study.…”

“…However, in situ scanning is not always feasible to integrate into existing autopsy protocols, including at the brain banks used in our study. Like the current paper, Alegro et al [4] also utilize 3D printing, but for a different purpose, to create customized containers to minimize brain distortion during formalin fixation. While the whole-brain mapping and in situ imaging are impressive advantages of the [4] pipeline, the amount of manual interaction in their pipeline is greater, requiring manual initialization for spline-based deformable registration for each histology slide.…”

“…Tward et al [72] also use deep learning to generate tangle heat maps; however they use a simpler patch classification strategy (assigning each patch a tangle/non-tangle label, as opposed to weakly supervised learning used in this paper) and do not evaluate resulting heat maps against manual tangle counts or semi-quantitative scores. Given that previous tau 3D mapping efforts [72, 4] were applied to only one or two specimens, the current work is the first to analyze histologically-derived 3D maps of tau pathology in a group template space and derive average and summary maps of NFT pathology burden (Fig. 6).…”

“…Like the current paper, Alegro et al [4] also utilize 3D printing, but for a different purpose, to create customized containers to minimize brain distortion during formalin fixation. While the whole-brain mapping and in situ imaging are impressive advantages of the [4] pipeline, the amount of manual interaction in their pipeline is greater, requiring manual initialization for spline-based deformable registration for each histology slide. Tward et al [72] describe a new deformable registration approach for aligning serial anti-tau immunohistochemistry slides to ex vivo MRI without a need for intensity transfer, and implement it for several tissue blocks from one brain donor.…”