Advancing Neurodegenerative Disease Diagnosis with Artificial Intelligence
Pedro Pinheiro-Chagas, Professor
UC San Francisco
Applications for Fall 2024 are closed for this project.
The UCSF Memory and Aging Center (MAC) Co-pilot project is an innovative research initiative aimed at revolutionizing the diagnosis of neurodegenerative diseases (NDD), such as Alzheimer's disease and Frontotemporal dementia, through the integration of advanced Large Language Models (LLMs). This project is particularly significant due to the high prevalence of NDDs and the emergence of new treatments necessitating early and accurate diagnosis. The core methodology of MAC Co-pilot involves the comprehensive integration and standardization of diverse patient data, including medical history, neuropsychological testing, and neuroimaging exams. The project utilizes LLMs to process, summarize, and analyze this data, aiming to replicate the clinical reasoning process of medical professionals. This approach is expected to significantly enhance diagnostic efficiency and accuracy, particularly in primary care settings, where it could have the most impact. With a sample of thousands of patients from the UCSF Memory and Aging Center, the project explores various dimensions of patient data. The hypothesis is that LLMs will demonstrate high diagnostic accuracy for common NDD syndromes and varying accuracy for rarer conditions. Furthermore, the project anticipates an improvement in the model’s performance when fine-tuned with additional neuropathology data. A crucial aspect of this research is the commitment to maintaining the confidentiality and integrity of patient data, ensuring compliance with the highest standards of data security and privacy. Finally, through a series of adversarial attacks, we will investigate the most important features used by the model, providing insights into the multidimensional nature of these complex diseases. This critical analysis aims to identify and strengthen the model's predictive robustness, ensuring that the artificial intelligent (AI) system can withstand potential data perturbations and biases, ultimately leading to more resilient and insightful diagnostic tools.
Role: Successful candidates will play a role in multiple aspects of the project. Responsibilities will encompass data integration and standardization, which includes importing data from diverse sources, and data cleaning to ensure the dataset’s integrity. You will also assist with the AI aspects of the tool, such as processing data, fine-tuning the model, and creating summary reports. An important part of your role will be ensuring the secure and ethical handling of sensitive patient data, adhering to the HIPAA standards. This position offers a unique opportunity to leverage AI technology within a healthcare context.
Qualifications: Ideal candidates should have a background in computer science, neuroscience, or a related field, with a good understanding of Python and some knowledge of machine learning. Though not strictly necessary, familiarity with AI is beneficial. Most importantly, we value a passion for leveraging technology to improve healthcare.
Hours: to be negotiated
Related website: https://profiles.ucsf.edu/pedro.pinheirochagas
Related website: https://mathcognition.ucsf.edu/