Presentation by David Gutman, Assistant Professor, Biomedical Informatics, Emory University School of Medicine

Deep Learning Workflow for Scalable Detection of Tau Inclusions in Digital Pathology

Share this article:

A novel computational workflow to train annotator specific models and ensemble models.

Pathological hallmarks of Alzheimer’s disease and related dementias include intraneuronal inclusions of hyperphosphorylated tau. Manual detection and characterization of these and similar markers in glass slides or digital slides is not currently done at scale, due to the large number of them and the high degree of inter-rater disagreement in defining them visually. In this work we used a digital workflow for collecting annotations of tau inclusions to measure inter-rater agreement between neuropathology experts and novice participants. A novel computational workflow was then implemented to train annotator specific models and ensemble models. Using state-of-the-art object detection models, we show that these models approach consistent detection of tau inclusions comparable to human annotators, while adding scalability and consistency.