Reproduce “Can Coding Agents Build Robust Baselines? A Skill-Based Approach for Automating the Medical Imaging Model-Development Pipeline” 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
Eugenia Moris, José Ignacio Orlando
A structured AI Scientist workflow automates the deep learning model development pipeline for medical imaging applications, integrating literature-guided reasoning, code generation, and hypothesis-driven experimentation. The system achieved competitive performance on multiple public benchmarks, securing 6th place out of 15 teams on PUMA and 31st out of 125 submissions on MILK10k, while improving MIDOG25 F1 score from 0.757 to 0.855.
Key insights
- While literature-guided planning provides a robust starting point, systematic experimentation is critical for adapting solutions to specific dataset characteristics and achieving optimal performance.
- An 'experiment tree' that records both accepted and rejected experimental evidence effectively guides future iterations, preventing re-exploration of unsuccessful configurations.
- The workflow demonstrates generalizability across diverse medical imaging tasks (segmentation, classification, detection) and produces models robust to domain variability without explicit adaptation strategies.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.