Research map / Detection and segmentation / Open-set and few-shot detection
RadGPT: Constructing 3D Image-Text Tumor Datasets
Pedro R.A.S. Bassi, Mehmet Can Yavuz, Ibrahim Ethem Hamamci, Sezgin Er, Xiaoxi Chen, Wenxuan Li, Bjoern Menze, Sergio Decherchi, Andrea Cavalli, Kang Wang, Yang Yang, Alan Yuille and 1 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| generate reports from tumor segmentation (RadGPT) instead of end-to-end VLM | report generation · UCSF external (OOD) kidney tumor >2cm | sensitivity (%): 9.2 -> 97.3 | Improved |
| generate reports from tumor segmentation instead of end-to-end | report generation · UCSF external (OOD) liver tumor >2cm | sensitivity (%): 19.2 -> 89.4 | Improved |
| generate reports from tumor segmentation instead of end-to-end | report generation · AbdomenAtlas 3.0 test (IID) pancreatic tumor >2cm | sensitivity (%): 51.9 -> 81.5 | Improved |
| generate reports from tumor segmentation instead of end-to-end | report generation · AbdomenAtlas 3.0 test (IID) liver tumor >2cm | specificity (%): 95.9 -> 64.4 | Hurt |
| generate reports from tumor segmentation instead of end-to-end | report generation · UCSF external (OOD) pancreatic tumor <=2cm | sensitivity (%): 27.5 -> 76.9 | Improved |
| adapt structured reports to hospital narrative style (RadGPT-N vs RadGPT-S) (ablation) | RadGPT · UCSF external (OOD) text metrics | ROUGE-1: 13.56 -> 22.77 | Improved |
| narrative style versus structured style (ablation) | RadGPT · UCSF external (OOD) text metrics | BLEU: 0.72 -> 0.74 | Improved |
| narrative style versus structured style (ablation) | RadGPT · UCSF external (OOD) text metrics | RadGraph-F1: 4.15 -> 2.92 | Hurt |
| train baseline on AbdomenAtlas 3.0 reports (CT2Rep) and test OOD | CT2Rep · UCSF external (OOD) liver tumor >2cm | sensitivity (%): 49.2 -> 0.0 | Hurt |
| use RadGPT structured reports instead of human-made reports | report content · 90 UCSF CTs | volume and HU reported: 0% -> 100% | Improved |
| use RadGPT structured reports instead of human-made reports | report content · 90 UCSF CTs | all tumor diameters reported: 63% -> 100% | Improved |
| LLM labeler for tumor presence (zero-shot Llama 3.1 70B) | diagnostic evaluation · 447 radiologist-checked reports | accuracy: 96% | Improved |
Closest papers by meaning
- Scaling Tumor Segmentation: Best Lessons from Real and Synthetic Data (ICCV 2025 · similarity 0.88)
- CAT: Coordinating Anatomical-Textual Prompts for Multi-Organ and Tumor Segmentation (NeurIPS 2024 · similarity 0.86)
- "Why Is There a Tumor?": Tell Me the Reason, Show Me the Evidence (ICML 2025 · similarity 0.86)
- Towards Generalizable Tumor Synthesis (CVPR 2024 · similarity 0.85)
- PanTS: The Pancreatic Tumor Segmentation Dataset (NeurIPS 2025 · similarity 0.85)
- Large-scale and Fine-grained Vision-language Pre-training for Enhanced CT Image Understanding (ICLR 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
