Artificial Intelligence-Powered Quantification of Flortaucipir PET for Detecting Tau Pathology.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Alzheimer's Disease Research

Background:

• Alzheimer's disease (AD) diagnosis and monitoring often rely on tau positron emission tomography (PET) imaging.
• Quantifying tau PET uptake typically requires structural magnetic resonance imaging (MR), increasing costs and complexity.
• Earlier and more accessible methods for tracking tau pathology are needed for timely intervention.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI)-powered approach for efficient tau PET quantification without structural MR.
• To assess the clinical utility of AI-derived tau PET measures for early AD diagnosis and progression monitoring.

Main Methods:

• A deep neural network model was implemented for normalizing ¹⁸F-AV1451 (tau) PET images, utilizing transfer learning from amyloid PET models.
• An MR-free pipeline for tau PET quantification was established and validated using the Alzheimer's Disease Neuroimaging Initiative dataset.
• Correlations between AI-derived tau uptake, cognitive measures (AD stage, memory performance), and longitudinal cognitive decline were analyzed.

Main Results:

• The AI pipeline demonstrated high performance, with intraclass correlation coefficients > 0.97 compared to MR-based quantification.
• Significant correlations were found between metatemporal tau deposition and cognitive scores (MMSE, MoCA).
• Elevated tau PET uptake in specific regions predicted future cognitive decline, highlighting prognostic value.

Conclusions:

• The AI-powered pipeline enhances tau PET accessibility by reducing costs and simplifying quantification, eliminating the need for MR.
• The developed method provides cognitively relevant outcome measures for early AD detection and monitoring.
• This approach supports personalized treatment strategies by enabling precise tracking of AD biomarkers.