Research
At the NeuroAI Lab, we bridge powerful AI with cognitive neuroscience to tackle two interrelated challenges: revolutionizing the discovery, diagnosis and care of neurodegenerative diseases and elucidating the neural substrates of human symbolic cognition. Check out our perspective paper on Integrating Generative Artificial Intelligence in ADRD: A Roadmap for Streamlining Diagnosis and Care in Neurodegenerative Diseases
- Agentic AI systems to reason about vast multimodal datasets, accelerating disease discovery.
- Copilot decision support systems to scale specialist-level clinical and neuropathological diagnoses.
- AI voice agents to conduct clinical assessments, streamlining data collection, monitoring and care.
Agentic AI systems to reason about vast multimodal datasets, accelerating disease discovery:
This project aims to characterize the clinical and biological heterogeneity of Alzheimer’s disease (AD) and frontotemporal dementia (FTD), by leveraging state-of-the-art AI and traditional machine learning algorithms to analyze extensive electronic medical record (EMR) data. By combining structured EMR data (diagnoses, labs, medications, imaging) with unstructured information from millions of clinical notes, parsed with LLMs, we aim to identify clinical subtypes, key symptoms, and risk factors that inform disease onset, progression, and differential diagnosis. The project further leverages the UCSF MAC’s rich longitudinal cohort to validate data-driven discoveries. Ultimately, this work aims to develop clinically actionable models and decision-support systems that enhance early diagnosis and guide individualized care. See our preprint here: Characterizing Dementia Phenotypes from Unstructured EHR Notes with Generative AI and Interpretable Machine Learning. Main collaborators: Alice Tang, PhD and Marina Sirota, PhD.
Copilot decision support systems to scale specialist-level clinical and neuropathological diagnoses:
This project aims to transform neurodegenerative disease diagnosis through the Memory and Aging Center Co-pilot, an AI-powered decision support system that integrates multimodal patient data—including unstructured clinical notes, neuropsychological assessments, and structural MRI scans—via a scientifically grounded, modular multi-agent architecture. The system leverages advanced generative AI paired with discriminative machine learning classifiers trained on a large, deeply phenotyped cohort to deliver clinical and neuropathological diagnosis at specialist-level accuracy and at a fraction of the time typically required. The Co-pilot’s interactive chat-based interface enables clinicians to dynamically query both the model’s reasoning and underlying patient data, supporting collaborative decision-making. This project aims to empower clinicians with scalable specialist-level precision diagnostics, ultimately bridging gaps in early detection and equitable care for diverse populations. Our flagship project to augment clinical care was recently awarded a UCSF’s Catalyst Program (translational accelerator focused on advancing UCSF discoveries for patient benefit and commercialization). See our preprint here: Agentic Generative Artificial Intelligence System for Classification of Pathology-Confirmed Primary Progressive Aphasia Variants. Main collaborators: Bruce Miller, MD; Katherine Possin, PhD; Maria Luisa Gorno-Tempini, MD, PhD; Bill Seeley, MD, and Andreas Rauschecker, MD, PhD.
AI voice agents to conduct clinical assessments, streamlining data collection, monitoring and care:
This project centers on developing and validating the voice Clinical Dementia Rating (vCDR), an automated and scalable digital version of the Clinical Dementia Rating (CDR®)—a widely used, global benchmark for staging dementia severity and tracking cognitive and functional decline in Alzheimer’s disease and Related Dementias. The CDR is administered in-clinic through a semi-structured interview with both the participant and a study partner, whose combined perspectives are essential for assessing changes from an individual’s baseline abilities. The vCDR leverages generative AI (large language models, LLMs) to conduct these interviews, automate scoring, and optimize the analysis of additional features, including connected speech. The vCDR aims to provide a more sensitive, consistent, and accessible tool for detecting and monitoring dementia progression, validating it as a clinical trial endpoint, and laying the groundwork for inclusive, global dementia research and care. Main collaborators: Michael Weiner, PhD; Rachel Nosheny, PhD; Andrew Breithaupt, MD; and Jet Vonk, PhD.
░░ ░
░ ░░ ░░▒ ░░░
░▒▒▒░ ░ ░░░░
░ ░░ ░░▒░ ░
░░░░░ ░░░ ░░ ░
░ ░░░ ▒▒ ░░ ░░
░ ░░ ░ ░ ░░░░ ░▒▒
░ ░░ ░░ ░░░░░░ ░░▒░ ▒░░▒
░ ░ ░░ ░░▓▒ ░ ░░░▒▒░ ░ ░░ ░░░
░░░ ░░░ ▒░ ░ ░▒ ░░░░░░░▒▒░ ░░░░░░░
░▒▒░ ░░░ ░░▒▓▓░ ░ ░░ ░░ ░▒▓░ ░░▒░
░░░░ ▒░ ░▒░ ░░ ░░░ ░░ ░ ░░
░░▒▒░░░ ░▒░░░ ▒▒░▒░ ░▒▒▒▒▒▒ ░ ░▒
▒▒▒ ░▒▓▒░░░░ ▒▓▒▒▒░▒▒░ ░▒▒▓░ ░▒ ░░ ░▒▒░░
░░▒░ ░░░░░░░░░ ░▓▓ ░▒▒ ░▒▓ ░ ░ ░░ ░
░░ ░░░░ ▒▓▓▓█▓▒▒░░ ░ ░▒░ ░▒ ░░░▒▒▒▒▓▓▓ ░▒░ ░▒ ░ ░ ░░
░░ ░▒░ ░▒▒░▒▒ ░░▒▒░ ░░░ ░▒▒░░░ ▓█ ░░░░░░▒▒ ░░░░▒▒ ░░░ ░░ ░░ ▒▒░░ ░
░ ░░▒▒ ▒░ ░░░░ ░░▒▒░ ░▒░░░ ░░▒ ░▒▒▒░▒▓▒ ░░░▒▒░▒ ░░▒░░ ▒ ░
░▒▒▒░░░░ ░▒▒░ ░░▒▓▓▒ ▒▒▒░ ░░ ░░░▒▒░░▒░ ░░░▒▒▓▓░ ░░░▒░░░▒░ ░░ ░░ ░ ░░ ░ ░
░▒▓░ ▓▒░ ░░▒▒▓▓█▓ ░▒▓ ░░░░░░░▒█░░░▒▒░░░░░░▒▓▓░ ░░░▒░ ░░▓ ░░░░▒░ ░ ░▓░░ ░ ░░░░ ░
░░ ░▒▒░░ ░ ░░░░░▒▒▒▒████░ ░ ░▒ ░░░▒▒▒░░▒▓▒░░░ ░░░▒▒▒▓▒▒░▒░░░░░░▒▓░░░░░░ ░▒▓▓▓░░░▒░░░░░ ░░▒░
░░░░░░░ ░░░░░░░░░▒▒▒▓████░░░ ░ ░░ ░░▒░░░░░░░░░░▒▓▒░░░░░░░░▒▓▓ ░░░░░░▒▓░░░ ░ ░▒▒░ ░ ░ ░
▒░░▒▓░░░▒░░ ░░▒▒▒▓▓██▒░░░ ░░░ ░▒░ ░░░░▒░░░░░░░░▒▒▒▓█████▓▓▓██░ ░░░░░░▓▓░░ ░░░░░░░░░▓▒░ ░░ ░░ ░
▒░░░░░░░░▒░░░░░░░▓░ ░▒░░░ ░░▒░▒░░░▒░░░░░▒█▒░░░▒▒▒▓▓██▒░░░░▒░░░░ ░░░░░░░░▒▓▒▒░░░░░░▒░░░░░▒▓▒░░░ ░ ░ ░░
░░░░░▒░░░░▒▒▒░░▒▓▒▒░░ ▒▒░░░▒▓▒▒░░░▒▒░░░▒▒▓█░░░▒▒▓█░ ░░▒▒░ ░░░░░░░░░░▒█▒░ ▒▒░░▒▒░░░░░▒░▒▒░░░░░ ░ ░░
░░░▒▒▒▒▒░░░▓▓██▓░░░░░░░░░░░░▒█▓▓▓░░▒▒░░░▒▒▓▒░▒▒█▓░ ░░▒▒▒░░▒▓█▓░░ ░░░▒▓▒▒▓▓▓▓░░▒▓▒░░░░░▒▓▓▒▒▒▒░
░░░ ▒▒▒▓▓▓▓█▒░░▒░░░▒▒▒░░░░░▓████░░░░░░░▒▓▓▓▓█░ ░░░░░░▒▒▓█▓▒░░░░░░░░░░░░░▒▓▓░░░▒▓▒▒░░░ ░░
░░░░░░░░░░▒░ ░░▒░░▒▒▓▓▓▓▒▒▓████▓▓▓▒▒▒▒▓▓██▒░ ░░░░░░ ░░░░░░░░░░░░██▓▒░░░░░░░░░░░▒▓▒░░░░░▒▒▒░░░
░░░░░░░░░▒▒░░░░░░░░▒▒▒▓▓█████▓▒░░░░▒▓████▒░░░░░░░░░░░░░░░░░░░░░░░░▓█▓█▒░░░░░░░░░░░░▒░░▒░░ ▒▒▒░░░ ░░░░
░░▒░░░░░░▒▒░░░░▒▒▒▒▒▒▓▓██▒░░ ░░░░▒▓▒░ ░▒▒░░░░░░░░░▓███▓▓▒▒▒░░░░░▓▒░░░░░░░░░░░░░▒▓▒░░▒▒░░░░ ░▒▒▒▒░░
░░░▒░░░░░░▒▒▓▓▓▓▒▒▒▒▓▓▓░░ ░▒▒▒▒▒░░ ░░▒▒████▓▒░░░░░░░░░░░░░░▒███▓▓▓▓▓▓▓▓▒▓█▓▒▒░▒▒▒░░░░ ░░░░░ ░▒░
░░░░░░░░░▒▒▒▒▒▓▓▓░ ░░░░░▒▓▒ ░░░░ ░░░░░░▒▒▓▓░░░░░░░░░░░░░░░▒█▓▓░░░░░░░░░░░░░▒▒░░░░░░░░░ ░
░░░░░░▒▓█░ ░░░░░░░░░░░░░░▒▒▓▒░░░░▒▒▒▒▒▓▓▓▓▒▒░░▒░░░░░░░░░░░ ▓░░░░░░░▒▒▒░░
░░░░░░▒▓▓░░░░░░▒░░▒░░░▒▒░░░░▒▓ ░░░▓▒░░▒░░░░░░░░░░░░░▒▒▒▒░░░ ░░▒▒▒▒░░░░ ░░░░░
░░░░▒▒▒▓▒░░░░░░▒▒▒▒░░▒▒░░░░░░▓░░░░░░░▒▒▓▒░░░▓▓█▓▓▓▓▓▒░░░░░░ ▒▒░░░░░▒▒▒▒▒░░ ░░
░░░░░░░░▒░░░▒▒▒▒▒▒▓░░░░░░░░░░░▒░░░░░░░░░░░▒▒▒▒▒▒▒▒░░░░░░░░░ ░░▒░░░░░░░░░░░
░░░░░░░▒░░░░▒▒▓░░░▒░░░░░░░░░░▒░░░░░░░░░░░░░░░░░░░░░░░░░░ ░░░░░░░░░░░░░
░░░░░░░░░░░░ ░░░░░░▒▒░░░▒▒▒▒░░░▓▒░░░░▒▒▒▓▓▒░▒░░░░░▓█████▓▓▓▓▓▓▒░░░░░░░░░
▒▒░░░░░░░░░░░░░░░░░░▒▓▓▒░░░░▒░░░░░░░░░░░░░░░░▒▓▓█▓█▓▓▓▓▓▒▒▒▒▒▒▒▒░░▒░░░
░░░░░░░░▒░░░░░░▒▒▒▒▒▓▒░░▒░░░░░░░░ ▒▓████████████▓▓▓▒▒▒▒▒▒▒░▒▒▒▒░░░░░
▒░░░░░▒▒▒░░░░░░░▒▓░░░░░░░░ ███▓█▓▓▓▓████▓▒▒▒▒░░▒▒▒▒▒▓▒░░░░░
░░░▒▒░▒░░░░░▒▒░░░░ ▓▓▓▓█▓▓▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒░░░░░
░░▒░ ▒█▓▓▓▓▒▒▒▒▒▒▒▒▒▒░▒▒░░▒▒▒░
░ ▓▒▒▒▒▒▒▒▒░▒▒▒▒
▓██▓▒░
█▓▒
-->