Agentic Generative Artificial Intelligence System for Classification of Pathology-Confirmed Primary Progressive...

Area of Science:

• Neuroscience and Artificial Intelligence
• Diagnostic accuracy in neurodegenerative diseases

Background:

• Accurate diagnosis is crucial for neurodegenerative diseases, especially with new therapies.
• Primary Progressive Aphasia (PPA) diagnosis is complex due to varied presentations and data integration challenges.
• Generative AI offers potential for scalable diagnostic support in complex neurological conditions.

Purpose of the Study:

• To assess the diagnostic performance of an agentic generative AI system for classifying PPA.
• To evaluate AI's ability to determine PPA clinical syndromes and underlying pathologies.

Main Methods:

• A retrospective diagnostic validation study utilized a multi-agent generative AI architecture.
• The AI processed multimodal data: clinical notes, neuropsychological tests, and MRI scans from 54 PPA patients.
• The system performed open-ended and constrained classifications against gold-standard post-mortem diagnoses.

Main Results:

• The AI achieved 90.7% accuracy in identifying PPA cases in an open-ended setting.
• Classification accuracy for PPA variants was high: 100% for semantic (svPPA) and nonfluent (nfvPPA), 94.1% for logopenic (lvPPA).
• Neuropathological prediction accuracy reached 100% for FTLD-TDP type C and FTLD-4R tau, and 94.4% for Alzheimer's disease; the entire process took under 10 minutes.

Conclusions:

• The generative AI system demonstrated expert-level diagnostic performance for PPA, integrating multimodal data effectively.
• The AI's speed and accuracy suggest potential for expanding access to specialized diagnostic expertise.
• Further validation in diverse patient populations is recommended for broader clinical application.