CICI: IPAAI: A Traceable, Cost-Aware, and End-to-End Data Curation Framework for Medical AI

Medical artificial intelligence systems are only as trustworthy as the data used to train and validate them. In distributed medical cyberinfrastructure, data quality problems can arise during imaging acquisition, electronic health record extraction, expert annotation, data linkage, and preprocessing. These problems often become visible only after deployment, when a model fails for a patient subgroup, produces unexplained errors, or performs poorly in a new clinical setting. Current curation practice depends on manual investigation that is slow, difficult to reproduce, and hard to audit across institutions. This project develops a traceable, cost-aware, and end-to-end framework for data quality assurance in medical artificial intelligence. The project creates algorithms that trace model failures to responsible data, diagnose likely root causes, and select remediation actions under practical constraints such as expert time, data scarcity, and computing cost. It builds provenance infrastructure that records actionable and queryable lineage across heterogeneous medical pipelines and training runs. It also integrates these capabilities into a curation system with human review for high-stakes decisions, using oncology workflows involving medical imaging, electronic health records, and expert annotations as a testbed. The expected results will help healthcare researchers and institutions build more reliable, auditable, and reusable medical artificial intelligence datasets. The project will also produce open-source tools, benchmark resources, educational materials, workshops, student competitions, and public outreach activities that broaden participation in trustworthy medical artificial intelligence. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria. NSF Award ID: 2613761 | Program: 01002627DB NSF RESEARCH & RELATED ACTIVIT | Principal Investigator: Ruoxi Jia | Institution: Virginia Polytechnic Institute and State University, BLACKSBURG, VA | Award Amount: $899,998 View on NSF Award Search: https://www.nsf.gov/awardsearch/show-award/?AWD_ID=2613761 View on Research.gov: https://www.research.gov/awardapi-service/v1/awards/2613761.html