Penn Image Computing and Science Laboratory

Deep learning measures of tau and TDP-43 explain medial temporal lobe atrophy

The medial temporal lobe (MTL) is a hotspot for neuropathology, and MTL atrophy is widely used as a biomarker of cognitive decline. Because several proteins accumulate there at once, it has been hard to say which pathology drives the atrophy.

In a study in Acta Neuropathologica, Amanda Denning, Paul Yushkevich and colleagues developed two deep learning algorithms that measure phosphorylated tau (p-tau) and TDP-43 pathology on digitized histology sections with annotated MTL subregions. The focus was on Alzheimer’s disease and limbic-predominant age-related TDP-43 encephalopathy (LATE).

Key findings:

The results argue for quantitative neuropathology in future studies of how pathology relates to brain structure.

Annotated hippocampal histology sections and deep learning WildCat activation maps for p-tau and TDP-43. Figure 2 from Denning et al., Acta Neuropathologica 2024 , CC BY 4.0 https: creativecommons.org licenses by 4.0 .
Annotated hippocampal histology sections and deep learning (WildCat) activation maps for p-tau and TDP-43. Figure 2 from Denning et al., Acta Neuropathologica (2024), CC BY 4.0.

Other PICSL contributors include Ranjit Ittyerah, Lisa Levorse, Chinmayee Athalye, Sadhana Ravikumar, Mengjin Dong, Yue Li, Pulkit Khandelwal, Laura Wisse and Sandhitsu Das.

Reference: Denning AE, Ittyerah R, Levorse LM, et al., Yushkevich PA. Association of quantitative histopathology measurements with antemortem medial temporal lobe cortical thickness in the Alzheimer’s disease continuum. Acta Neuropathologica. 2024. doi:10.1007/s00401-024-02789-9