Fig. 2
From: Artificial intelligence in neuropathology: deep learning-based assessment of tauopathy
Schematic overview of data annotation and deep learning pipeline. NFT are annotated using whole slide images, marking rectangular areas with complete annotations of NFT objects to give a visual context of NFT against the background. Image patches are extracted for network training that generates pixel-wise segmentations for NFT and background. Performance is determined using a separate novel set of images (test set) by comparing expert annotation with the trained network. Quantitative evaluation on unseen test slides provides a characterization of the behavior of the trained network in terms of robustness, reliability, and reproducibility
