Artificial intelligence-derived neurofibrillary tangle burden is associated with antemortem cognitive impairment

• DOI: 10.1186/s40478-022-01457-x
• Publication Date: 2022-10-31T10:02:56Z
• Authors: Gabriel A. Marx; Daniel G. Koenigs...; Andrew T. McKenzie; Justin Kauffman; Russell W. Hanson; Kristen Whitney; Maxim Signaevsky; Marcel Prastawa; Megan A. Iida; Charles L. White; Jamie M. Walker; Timothy E. Richar...; John Koll; Gerardo Fernandez; Jack Zeineh; Carlos Cordon-Cardo; John F. Crary; Kurt Farrell
• Keywords: Tauopathy; Tangle; Entorhinal cortex; Neurofibrillary tangle

Abstract Tauopathies are a category of neurodegenerative diseases characterized by the presence abnormal tau protein-containing neurofibrillary tangles (NFTs). NFTs universally observed in aging, occurring with or without concomitant accumulation amyloid-beta peptide (Aβ) plaques that typifies Alzheimer disease (AD), most common tauopathy. Primary age-related tauopathy (PART) is an Aβ-independent process affects medial temporal lobe both cognitively normal and impaired subjects. Determinants symptomology subjects PART poorly understood require clinicopathologic correlation; however, classical approaches to staging pathology have limited quantitative reproducibility. As such, there critical need for unbiased methods quantitatively analyze on histological level. Artificial intelligence (AI)-based convolutional neural networks (CNNs) generate highly accurate precise computer vision assessments digitized slides, yielding novel histology metrics at scale. Here, we performed retrospective autopsy study large cohort ( n = 706) human post-mortem brain tissues from elderly individuals mild no Aβ (average age death 83.1 yr, range 55–110). We utilized CNN trained segment hippocampus sections immunohistochemically stained antisera recognizing hyperphosphorylated (p-tau), which yielded regional NFT counts, positive pixel density, as well graph-theory based metric measuring spatial distribution NFTs. found several AI-derived significantly predicted cognitive impairment proper entorhinal cortex p < 0.0001). When controlling age, counts still 0.04 cortex; overall). In contrast, Braak stage did not predict either age-adjusted unadjusted models. These findings support hypothesis burden correlates PART. Furthermore, our analysis strongly suggests provide powerful tool can deepen understanding role degeneration impairment.