Reproduce “Deep Learning Predicts Hip Fracture using Confounding Patient and Healthcare Variables” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Marcus A. Badgeley, John R. Zech, Luke Oakden-Rayner, Benjamin S. Glicksberg, Manway Liu, William Gale, Michael V. McConnell, Beth Percha and 2 more
Deep learning models predicting hip fractures from radiographs were found to heavily leverage confounding patient demographic and hospital process variables, with performance degrading to near-random when these variables were controlled. Explicitly integrating clinical data into a multimodal model improved fracture detection performance to an AUROC of 0.86.
Key insights
- Radiographs contain visual signatures of diverse patient traits and hospital process variables that deep learning models can detect.
- Deep learning models' apparent ability to detect hip fractures was largely driven by their capacity to detect and leverage these confounding variables.
- Explicitly integrating patient and hospital process data directly into multimodal AI models is more effective for fracture detection than relying solely on image data or a clinician interpreting separate model outputs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.