Reproduce “Deep Learning Models to Automate the Scoring of Hand Radiographs for Rheumatoid Arthritis” 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

Zhiyan Bo, Laura C. Coates, Bartlomiej W. Papiez

Researchers at the University of Oxford created an automated deep learning pipeline to predict Rheumatoid Arthritis (RA) severity by directly scoring hand radiographs using the van der Heijde modified Sharp (SvdH) method. This pipeline, leveraging ensemble learning and transfer learning, achieved continuous SvdH score prediction performance (PCC 0.925, MAE 12.57, RMSE 18.02) comparable to inter-rater agreement among experienced radiologists.

Key insights

  • Transfer learning from a related anatomical dataset (pediatric bone age X-rays) consistently improved model performance across all investigated CNN architectures for RA scoring.
  • Ensemble learning significantly boosted prediction accuracy, with the ensemble regressor achieving performance metrics for SvdH scoring that were remarkably close to, or slightly better than, human inter-rater agreement.
  • Direct, end-to-end prediction of SvdH scores from entire radiographs, without explicit joint localization, provides a robust and effective solution, simplifying the pipeline compared to multi-stage approaches.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.