Topic 433: A Federated Data Platform for Developing Unbiased Breast Imaging AI

This SBIR project aims to implement a federated data platform focused on a data system to enable mammography data to be published into a federated catalog from which collaborators can execute federated analyses.

It will enable hospitals and other clinical entities to make their data available for training and validation of artificial intelligence applications.

The data federation will include a racially, geographically, and socio economically diverse group of contributors that provide data representative of the US population. This diverse data set will provide AI developers with a resource to reduce bias in their algorithms.

The platform will include a data discovery portal, governance functions to manage access to the data, and a computational infrastructure to support federated learning and distributed algorithm validation.

The platform builds Flywheel’s existing product suite, which includes data management tools for extracting data from clinical information systems, anonymizing the data, and shepherding the data through a comprehensive data curation and quality review workflow.

Automated procedures extract rich metadata, which is then shared at an aggregate level to the Flywheel Data Exchange, from which federated data sets and analytic routines are managed.