Figure 5. Model-derived NFTDetector scores correlated with expert CERAD-like scores.

A) Representative WSIs spanning the four semi-quantitative stages of increasing NFT burden severity, ordered by CERAD-like scoring. Model NFT detections overlaid in green. B) Area-normalized NFTDetector scores on test set WSIs correlate with semi-quantitative ordinal categories. C) Area-normalized NFTDetector scores calculated on the full WSI dataset versus area-normalized counts for the same WSIs derived from the annotator labels. **: 0.001 < p <= 0.01 by Welch’s t-test. We used different assessment datasets in B) vs C) because the external test set lacks human NFT point annotations.