Acquiring Images with Harmonized Outcome Data for Crowdsourced Challenges

Abstract

Testing whether collaboratively engaging citizen-scientists across the world could improve algorithms for screening mammography requires access to a large dataset. Decreasing false positive recalls further demands for high-quality data on risk factors and cancer outcomes, linked with screening mammography images and harmonized data elements. Researchers from Kaiser Permanente Washington worked with SAGE Bionetworks to refine the challenge and sub-challenge questions and to extract necessary data elements to support this challenge using data acquired for clinical care and for research purposes. These efforts gave access to an unprecedented data size of 640K digital mammography images, corresponding to 86K women. Our experiences highlight the promise of these types of innovative projects, but also require evaluation to learn from our experiences for future collaborative endeavors.