Overview
Explore the intersection of artificial intelligence and radiology in this comprehensive lecture and primer from the Broad Institute. Delve into AI's ability to detect hidden patterns in X-ray images not visible to radiologists, presented by Judy Wawira Gichoya from Emory University School of Medicine. Examine examples of AI detecting "hidden signals" in X-rays, assess model generalizability, and discover a research roadmap for harnessing AI capabilities in patient care. Then, investigate group fairness in chest X-ray diagnosis with Haoran Zhang from MIT, analyzing deep learning models through the lens of algorithmic fairness definitions. Discuss the potential consequences of fairness interventions, explore spurious correlations, and question appropriate fairness definitions in clinical contexts. Gain insights into AI's impact on radiology, fairness considerations, and the future of medical imaging technology.
Syllabus
Intro
AI in clinical care
Fairness
Definitions
Roadmap
Blueprints
Are existing classifiers group fair
Are Texas radiologists group fair
Binary production
Calibration
Histogram
Efficiency
Risk Score
Equivalence
Summary
Minimax critical awareness
Label bias
Shortcut learning
Shortcut learning in real world
Can race be a shortcut
The tradeoff
Practical suggestions
Welcome back
Why Imaging
Race
Data
Results
Taught by
Broad Institute