INDEX
Imaging Data Exchange
The Big Question
What if high-quality and geographically representative imaging data were easily available for medical AI development?
The Problem
A large amount of imaging data is necessary to train artificial intelligence (AI) and machine learning (ML) algorithms. Such algorithms can help radiologists and pathologists make faster and better diagnostic decisions. Unfortunately, the teams that develop these algorithms often don’t have access to sufficient high-quality and representative medical imaging data to train their AI/ML models well. If the algorithms they create are biased or inaccurate, they won’t be suitable for clinical use.
The Solution
The ImagiNg Data EXchange (INDEX) program aims to create a platform that seamlessly links data providers, data users, and service providers with high quality images that will enable robust, trustworthy AI tool development for pathology and radiology. The program seeks to increase the number, type, and quality of images available for ML models, as well as boost geographic, racial, and ethnic diversity of images. INDEX also asks performers to develop tools to standardize imaging data and to train ML models in cloud environments. ARPA-H and the U.S. Food and Drug Administration (FDA) are collaborating on reducing barriers in obtaining data needed for INDEX. FDA’s Office of Science and Engineering Laboratories (OSEL) will develop regulatory science toolkits (RSTks) to evaluate data quality and usage. Importantly, the tools developed using INDEX will help doctors read medical images quicker and with higher confidence, reducing physician burnout and getting patients to a diagnosis – and potential treatment – faster.
Why ARPA-H?
ARPA-H can bring together stakeholders from a wide ecosystem – including federal regulators, data providers, health care providers, transition partners, and ML developers – to address the technical challenges and enable faster and better medical AI algorithm development.
Program Manager

Ileana Hancu, Ph.D.
Solicitation
Closed.