Radiology AI Competitions: From Challenge Design to Clinical Impact
Radiology-focused machine-learning competitions have emerged as powerful engines for innovation, providing curated datasets, objective benchmarks, and a global forum for testing ideas at scale. This session traces the complete life cycle of a successful challenge from designing an actionable clinical question and assembling multi-institutional, expertly annotated data to selecting fair evaluation metrics, orchestrating day-to-day logistics, and sustaining participant engagement. We will share practical guidance on safeguarding patient privacy, ensuring label quality, preventing submission homogeneity and “shortcut” solutions, and maintaining transparency. Finally, we will discuss how winning algorithms are validated, and can be integrated into clinical workflows, highlighting real-world examples of tools that progressed from leaderboard to bedside. A live Q&A will explore challenges, pitfalls, external validation, and potential clinical deployment.
Objectives
Map the end-to-end workflow of a radiology AI competition including clinical problem selection, dataset curation, labeling methodology, metric design, and platform logistics
Identify common pitfalls and mitigation strategies related to data privacy, annotation accuracy, model overfitting, and generalization
Describe measurable outcomes such as open datasets, reference algorithms, and scientific insights, and how they accelerate regulatory clearance and clinical adoption
Outline post-competition pathways for external validation, generalization studies, licensing, and integration of top-performing models into routine practice
Speakers
Tara Retson, MD, PhD
Breast Radiologist and Assistant Professor
UC San Diego Health
Jeff Rudie, MD, PhD
Neuroradiologist, Scripps Clinic Medical Group
Adjunct Assistant Professor of Radiology, UC San Diego