Age-informed, attention-based weakly supervised learning for neuropathological image assessment.

Area of Science:

• Digital neuropathology
• Artificial intelligence in medicine
• Neurodegenerative disease research

Background:

• Diagnosing neurodegenerative disorders (NDs), including Chronic Traumatic Encephalopathy (CTE), is challenging due to subtle pathological changes.
• Current histopathological analysis is labor-intensive, subjective, and may miss critical alterations.
• Accurate quantification of tau pathology, a hallmark of CTE, is essential for diagnosis.

Purpose of the Study:

• To develop an automated, age-informed, attention-based multiple instance learning pipeline for predicting AT8 density (p-tau aggregation) in CTE.
• To generate interpretable attention maps for visualizing structural changes associated with tau pathology.
• To establish quantitative benchmarks for evaluating foundation models in neuropathology.

Main Methods:

• Utilized Luxol Fast Blue and Hematoxylin & Eosin stained whole-slide images.
• Implemented an attention-based multiple instance learning framework incorporating patient age.
• Developed quantitative metrics to assess foundation model performance, including attention map smoothness, faithfulness, and robustness.

Main Results:

• The age-informed model accurately predicts AT8 density, identifying critical pathological regions.
• Generated attention maps highlight structural changes linked to tau pathology, enhancing interpretability.
• The developed evaluation procedures provide a robust framework for assessing foundation models in neuropathological image analysis.

Conclusions:

• The proposed AI pipeline enables scalable, automated whole-slide image analysis for neurodegenerative disorder diagnosis.
• Incorporating patient age significantly improves predictive accuracy and contextual understanding of tau pathology.
• This approach supports earlier, more precise diagnoses of NDs and identifies subtle pathological markers for potential clinical imaging applications.