Reproduce “ReXGroundingCT: A 3D Chest CT Dataset for Segmentation of Findings from Free-Text Reports” 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
Mohammed Baharoon, Luyang Luo, Michael Moritz, Abhinav Kumar, Sung Eun Kim, Xiaoman Zhang, Miao Zhu, Mahmoud Hussain Alabbad and 17 more
Researchers from Harvard Medical School created ReXGroundingCT, the first publicly available 3D chest CT dataset that meticulously links free-text radiology findings to precise pixel-level 3D segmentations. This dataset, built through a rigorous process involving GPT-4 for report standardization and expert radiologist annotations, enabled fine-tuned models to achieve a Global Dice score of 0.209 for text-prompted 3D medical segmentation.
Key insights
- Large Language Models can be effectively integrated into data curation pipelines to standardize and extract clinical text, accelerating the preparation of complex medical datasets under expert oversight.
- Meticulous manual 3D annotation and quality control by medical experts are essential to establish high-fidelity links between nuanced free-text descriptions and pixel-level segmentations.
- The dataset supports the development of AI systems capable of "grounded radiology report generation," where textual descriptions are explicitly tied to specific spatial regions within images, enhancing explainability.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.