Reproduce “CheXpert: A Large Chest Radiograph Dataset with Uncertainty Labels and Expert Comparison” 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
Jeremy Irvin, Pranav Rajpurkar, Michael Ko, Yifan Yu, Silviana Ciurea-Ilcus, Chris Chute, Henrik Marklund, Behzad Haghgoo and 12 more
Stanford University researchers developed CheXpert, a large dataset of over 224,000 chest radiographs with labels for 14 observations, uniquely addressing inherent uncertainty in radiological findings. Their deep learning models achieved performance comparable to or exceeding individual radiologists for several key diagnostic tasks, including an AUC of 0.97 for Pleural Effusion, providing a benchmark for medical image interpretation.
Key insights
- No single method for handling uncertain labels is universally optimal; the best approach can vary by specific observation.
- Deep learning models, specifically DenseNet-121, can achieve performance comparable to or exceeding individual radiologists for several chest radiograph interpretation tasks.
- Model visualizations (Grad-CAM) demonstrate that the models focus on anatomically relevant regions, aligning with radiological interpretation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.