Fig. 1

Overview of project data contribution. Each neuropathology case contained four regions immunohistochemically stained for tau pathology (top). All cases included WSIs from the posterior hippocampus, temporal cortex, and occipital cortex. Emory cases included the amygdala while UC Davis cases used the anterior hippocampus instead. WSIs from the Emory-Train cohort included one or more ROIs that were annotated by either multiple annotators for inter-annotator-agreement analysis, a single annotator, or not annotated (middle, left). Not all WSIs contained ROIs in this cohort however. Annotations were collected as single pixels (circles for visualization) and converted to box annotations using watershed approach and manual quality check step (middle, right). The inference workflow was run on all WSIs to predict NFT subtypes (Pre-NFT or iNFTs), followed by extraction of imaging features used to train and test machine learning models for predicting Braak stages for neuropathology cases. Red circles/boxes represent iNFT annotations, blue circles/boxes represent Pre-NFT annotations. Two Emory cases contained two WSIs for the posterior hippocampus, one from each side of the brain (hemispheres). Overview of project cohorts and specific uses for different cohorts