Elamur Research⌕
☰
home / papers / 2026 / CPR-RAG: Clinical Prior-Regularized Retrieval for Anatomy-Aware 3D CT Report Generation

CPR-RAG: Clinical Prior-Regularized Retrieval for Anatomy-Aware 3D CT Report Generation

Paper · 2026 · Sungkyu Yang, Kang-Min Kim and Mansu Kim

Summary

Writing radiology reports for 3D chest CT is difficult. Small lesions get lost when a whole volume is compressed into a few visual tokens, and generators tend to fall back on generic "normal" templates. Standard retrieval-augmented generation helps only partly. Cases that look alike may differ clinically, and whole retrieved reports bring in large amounts of normal text about other organs.

CPR-RAG is a plug-and-play retrieval layer for existing 3D VLM report generators, and it works in three steps:

  • Organ-specific embeddings: learnable queries cross-attend over tokens from a frozen vision encoder to produce one embedding per organ. These embeddings are trained with per-organ multi-label disease classifiers.
  • Re-ranking: organ-level candidates retrieved from the training corpus are re-ranked using disease co-occurrence statistics computed from that corpus.
  • Context filtering: boilerplate normal sentences are removed before the context reaches the generator.

On RadGenome-ChestCT, adding CPR-RAG to CT2Rep, M3D and RadFM improves clinical efficacy, with large gains on rare findings. Human raters also preferred its reports. This matters because it attacks missed diagnoses, the most dangerous error in screening.

Key points

  • Setup: the framework is model-agnostic and covers five anatomical regions: Heart, Lung, Mediastinum, Pleura and Trachea/Bronchi. It is evaluated on RadGenome-ChestCT with the CT2Rep, M3D and RadFM backbones.
  • Headline result: clinical F1 rises by +16.36 over baselines. The absolute score (about 32%) is still too low for autonomous clinical use.
  • Fewer missed diagnoses: RadFM labels 47.8% of cases as fully negative, against a true rate of 12.3%. Its false-negative rate drops from 42.3% to 1.7% with CPR, at the cost of more false alarms.
  • Rare pathologies (≤10% prevalence): Micro F1 rises from 6.94% to 25.14% for M3D and from 3.92% to 21.08% for RadFM. For rare disease pairs (<1% joint prevalence), all baselines score zero, while the CPR versions reach 1.50–2.68%.
  • Limitations: the data come from a single institution, the report section order is fixed, and the comorbidity prior models only pairwise relations. The code is released at github.com/aimed-gist/cpr-rag.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksComputer Vision and Pattern Recognition · 2,869 worksMedicine · 241 worksHealth Informatics · 184 worksRadiology, Nuclear Medicine and Imaging · 3 works

🧭 Topics 1Research subjects and languages the work is about.Information retrieval & search · 473 works

Cite

Cite as Sungkyu Yang, Kang-Min Kim and Mansu Kim. 2026. CPR-RAG: Clinical Prior-Regularized Retrieval for Anatomy-Aware 3D CT Report Generation. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-long.1411

@inproceedings{yang2026cpr,
  title = {{CPR-RAG: Clinical Prior-Regularized Retrieval for Anatomy-Aware 3D CT Report Generation}},
  author = {Sungkyu Yang, Kang-Min Kim and Mansu Kim},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-long.1411},
  url = {https://doi.org/10.18653/v1/2026.acl-long.1411}
}

References

38 references, 13 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Fan Bai, Yuxin Du, Tiejun Huang, Max Q-H Meng, and Bo Zhao. 2024. M3d: Advancing 3d medical image analysis with multi-modal large language models. arXiv preprint arXiv:2404.00578. \#\# Acknowledgments We thank the anonymous reviewers for their helpful comments. This research was supported by the Natio… — [arXiv:2404.00578](https://arxiv.org/abs/2404.00578) find
  2. Satanjeev Banerjee and Alon Lavie. 2005. Meteor: An automatic metric for mt evaluation with improved correlation with human judgments. In Proceedings of the ACL workshop on intrinsic and extrinsic evaluation measures for machine translation and/or summarization, pages 65–72.
  3. Zhihong Chen, Yan Song, Tsung-Hui Chang, and Xiaojun Wan. 2020. Generating radiology reports via memory-driven transformer. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1439–1449.
  4. Mark Endo, Rayan Krishnan, Viswesh Krishna, Andrew Y. Ng, and Pranav Rajpurkar. 2021. Retrieval-based chest x-ray report generation using a pre-trained contrastive language-image model. In Proceedings of Machine Learning Research, volume 158 of Proceedings of the 6th Machine Learning for Health Conf… find
  5. Andre Esteva, Alexandre Robicquet, Bharath Ramsundar, Volodymyr Kuleshov, Mark DePristo, Katherine Chou, Claire Cui, Greg Corrado, Sebastian Thrun, and Jeff Dean. 2019. A guide to deep learning in healthcare. Nature Medicine, 25(1):24–29. find
  6. Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, Angela Fan, and 1 others. 2024. The llama 3 herd of models. arXiv preprint arXiv:2407.21783.
  7. Ibrahim Ethem Hamamci, Sezgin Er, and Bjoern Menze. 2024a. Ct2rep: Automated radiology report generation for 3d medical imaging. In Medical Image Computing and Computer Assisted Intervention – MIC-CAI 2024, pages 476–486. Springer. find
  8. Ibrahim Ethem Hamamci, Sezgin Er, Enis Simsar, Alperen F Tabak, Baris Atac, Ayse Betul Kose, Ecem Masazade, Berke Sevgi, Zeynep Sevgi, Gozde Unal, and 1 others. 2024b. Ct-rate: A high-fidelity 3d chest ct-report dataset for multi-modal foundation model pretraining. arXiv preprint arXiv:2403.17834. — [arXiv:2403.17834](https://arxiv.org/abs/2403.17834) find
  9. Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2022. Lora: Low-rank adaptation of large language models. In International Conference on Learning Representations.
  10. Jeremy Irvin, Pranav Rajpurkar, Michael Ko, Yifan Yu, Silviana Ciurea-Ilcus, Chris Chute, Henrik Marklund, Behzad Haghgoo, Robyn Ball, Katie Shpanskaya, and 1 others. 2019. Chexpert: A large chest radiograph dataset with uncertainty labels and expert comparison. In Proceedings of the AAAI Conference… find
  11. Ziwei Ji, Nayeon Lee, Rita Frieske, Tiezheng Yu, Dan Su, Yan Xu, Etsuko Ishii, Ye Jin Bang, Andrea Madotto, and Pascale Fung. 2023. Survey of hallucination in natural language generation. ACM Computing Surveys, 55(12):1–38.
  12. Jeff Johnson, Matthijs Douze, and Hervé Jégou. 2019. Billion-scale similarity search with gpus. IEEE Transactions on Big Data, 7(3):535–547.
  13. Seowoo Lee, Jiwon Youn, Hyungjin Kim, Mansu Kim, and Soon Ho Yoon. 2025. Cxr-llava: a multimodal large language model for interpreting chest x-ray images. European Radiology, pages 1–13. find
  14. Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, and 1 others. 2020. Retrieval-augmented generation for knowledge-intensive nlp tasks. In Advances in Neural Information Processing Systems (NeurI…
  15. Chin-Yew Lin. 2004. Rouge: A package for automatic evaluation of summaries. In Text summarization branches out, pages 74–81.
  16. Jingyang Lin, Yingda Xia, Jianpeng Zhang, Ke Yan, Le Lu, Jiebo Luo, and Ling Zhang. 2024. Ct-glip: 3d grounded language-image pretraining with ct scans and radiology reports for full-body scenarios. arXiv preprint arXiv:2404.15272. — [arXiv:2404.15272](https://arxiv.org/abs/2404.15272) find
  17. Geert Litjens, Thijs Kooi, Babak Ehteshami Bejnordi, Arnaud Arindra Adiyoso Setio, Francesco Ciompi, Mohsen Ghafoorian, Jeroen A. W. M. van der Laak, Bram van Ginneken, and Clara I. Sánchez. 2017. A survey on deep learning in medical image analysis. Medical Image Analysis, 42:60–88. find
  18. Fenglin Liu, Changchang Yin, Xian Wu, Shen Ge, Ping Zhang, and Xu Sun. 2021. Contrastive attention for automatic chest x-ray report generation. In Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 269–280. Association for Computational Linguistics. find
  19. Nelson F Liu, Kevin Lin, John Hewitt, Ashwin Paranjape, Michele Bevilacqua, Fabio Petroni, and Percy Liang. 2024. Lost in the middle: How language models use long contexts. Transactions of the Association for Computational Linguistics, 12:157–173.
  20. Ilya Loshchilov and Frank Hutter. 2019. Decoupled weight decay regularization. In International Conference on Learning Representations. find
  21. Yuren Mao, Wenyi Xu, Yuyang Qin, and Yunjun Gao. 2025. Ct-agent: A multimodal-llm agent for 3d ct radiology question answering. arXiv preprint arXiv:2505.16229. — [arXiv:2505.16229](https://arxiv.org/abs/2505.16229) find
  22. Yasuhide Miura, Yuhao Zhang, Emily Tsai, Curtis Langlotz, and Dan Jurafsky. 2021. Improving factual completeness and consistency of image-to-text radiology report generation. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human L… find
  23. OpenAI. 2023. Gpt-4 technical report. arXiv preprint arXiv:2303.08774.
  24. Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. Bleu: a method for automatic evaluation of machine translation. In Proceedings of the 40th annual meeting of the Association for Computational Linguistics, pages 311–318. Association for Computational Linguistics.
  25. Maithra Raghu, Chiyuan Zhang, Jon Kleinberg, and Samy Bengio. 2019. Transfusion: Understanding transfer learning for medical imaging. In Advances in Neural Information Processing Systems, volume 32. find
  26. Samyam Rajbhandari, Jeff Rasley, Olatunji Ruwase, and Yuxiong He. 2020. Zero: Memory optimizations toward training trillion parameter models. In SC20: International Conference for High Performance Computing, Networking, Storage and Analysis, pages 1– 16. IEEE. find
  27. Mercy Ranjit, Gopinath Ganapathy, Ranjit Manuel, and Tanuja Ganu. 2023. Retrieval augmented chest x-ray report generation using openai gpt models. In Proceedings of Machine Learning Research, volume 219 of Proceedings of the 8th Machine Learning for Health Conference (MLHC 2023), pages 650–666. PMLR… find
  28. Arun James Thirunavukarasu, Daniel Shu Wei Ting, Kallisvaran Elangovan, Laura Gutierrez, Ting Fang Tan, and David Shu Wei Ting. 2023. Large language models in medicine. Nature medicine, 29(8):1930– 1940. find
  29. Tao Tu, Shekoofeh Azizi, Danny Driess, Mike Schaekermann, Mohamed Amin, Pi-Chuan Chang, and 1 others. 2024. Towards generalist biomedical ai. NEJM AI, 1(3):AIoa2300138. find
  30. Wenhui Wang, Hangbo Bao, Li Dong, Johan Bjorck, Zhiliang Peng, Qiang Liu, Kriti Aggarwal, Owais Mohammed, Saksham Singhal, Subhojit Som, and 1 others. 2023. Image as a foreign language: Beit pretraining for all vision and vision-language tasks. In Proceedings of the IEEE/CVF Conference on Computer V… find
  31. Xiaosong Wang, Yifan Peng, Le Lu, Zhiyong Lu, Mo-hammadhadi Bagheri, and Ronald M Summers. 2017. Chestx-ray8: Hospital-scale chest x-ray database and benchmarks on weakly-supervised classification and localization of common thorax diseases. In Proceedings of the IEEE conference on computer vision an… find
  32. Chaoyi Wu, Xiaoman Zhang, Ya Zhang, Hui Hui, Yanfeng Wang, and Weidi Xie. 2025. Towards generalist foundation model for radiology by leveraging web-scale 2d&3d medical data. Nature Communications, 16(1):7866. find
  33. An Yan, Yu Wang He, and Chun-Nan McAuley, Julian andXHsu. 2022. Radbert: Adapting transformer-based language models to radiology. Radiology: Artificial Intelligence, 4(4):e210258. find
  34. Sixing Yan, William K. Cheung, Ivor W. Tsang, Keith Chiu, Terence M. Tong, Ka Chun Cheung, and Simon See. 2024. Ahive: Anatomy-aware hierarchical vision encoding for interactive radiology report retrieval. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), p… find
  35. Ori Yoran, Tomer Wolfson, Ori Ram, and Jonathan Berant. 2024. Making retrieval-augmented language models robust to irrelevant context. In International Conference on Learning Representations (ICLR).
  36. Tengfei Zhang, Ziheng Zhao, Chaoyi Wu, Xiao Zhou, Ya Zhang, Yanfeng Wang, and Weidi Xie. 2025. Radir: A scalable framework for multi-grained medical image retrieval via radiology report mining. In Medical Image Computing and Computer Assisted Intervention – MICCAI 2025, pages 508–518. Springer. find
  37. Xiaoman Zhang, Chaoyi Wu, Ziheng Zhao, Jiayu Lei, Ya Zhang, Yanfeng Wang, and Weidi Xie. 2024. Radgenome-chest ct: A grounded vision-language dataset for chest ct analysis. arXiv preprint arXiv:2404.16754. — [arXiv:2404.16754](https://arxiv.org/abs/2404.16754) find
  38. Yuhao Zhang, Derek Merck, Emily Tsai, Christopher D. Manning, and Curtis Langlotz. 2020. Optimizing the factual correctness of a summary: A study of summarizing radiology reports. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 5108–5120.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/yang-2026-cpr-rag-clinical-prior-regularized/