Reproduce “MURA: Large Dataset for Abnormality Detection in Musculoskeletal Radiographs” 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
Pranav Rajpurkar, Jeremy Irvin, Aarti Bagul, Daisy Ding, Tony Duan, Hershel Mehta, Brandon Yang, Kaylie Zhu and 6 more
MURA, a large dataset of 40,561 musculoskeletal radiographs, enables the development of deep learning models for abnormality detection, with a benchmarked DenseNet-169 model achieving an AUROC of 0.929 and performance comparable to radiologists for wrist and finger studies.
Key insights
- Large, high-quality, expertly-labeled datasets are essential for driving progress in deep learning for medical imaging, mirroring their impact in general computer vision.
- Deep learning models can achieve high overall abnormality detection performance (AUROC 0.929) but show variable agreement with human radiologists across different body regions.
- The use of Class Activation Maps (CAMs) demonstrated the model's ability to localize abnormalities, providing interpretability crucial for clinical acceptance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.