Reproduce “CheXbert: Combining Automatic Labelers and Expert Annotations for Accurate Radiology Report Labeling Using BERT” 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
Akshay Smit, Saahil Jain, Pranav Rajpurkar, Anuj Pareek, Andrew Y. Ng, Matthew P. Lungren
CheXbert, a BERT-based system, accurately extracts 14 common chest X-ray observations from radiology reports by employing a two-stage hybrid training approach. It achieved an F1 score of 0.798, representing a 0.055 F1 point improvement over previous rule-based methods and closely approximating radiologist-level accuracy.
Key insights
- A hybrid supervision strategy combining large-scale automatic labels with small-scale expert annotations yields superior performance in medical NLP.
- Domain-specific pre-training for language models significantly boosts effectiveness in specialized medical contexts.
- Data augmentation techniques like backtranslation are valuable for enhancing models trained with limited expert-annotated data.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.