Elamur Research⌕
☰
home / papers / 2026 / GraphRAG-Rad: Concept-Aware Radiology Report Generation via Latent Visual-Semantic Retrieval

GraphRAG-Rad: Concept-Aware Radiology Report Generation via Latent Visual-Semantic Retrieval

Paper · 2026 · Faezeh Safari, Hang Dong, Zeyu Fu et al.

Summary

Automatic radiology report generation systems often produce findings that sound credible but are clinically wrong. The authors (University of Exeter, EACL 2026 Student Research Workshop) present GraphRAG-Rad, which grounds generation in an external biomedical knowledge graph. Most RAG pipelines build text queries. Here, a Latent Visual-Semantic Retrieval (VSR) step instead maps the image's global ViT feature into the embedding space of PrimeKG nodes, whose names are encoded with PubMedBERT. It then pulls a top-K subgraph straight from the image. The subgraph is used in three ways:

  • A GAT encodes it, and the result steers attention over visual patches.
  • A recurrent module walks through it to imitate multi-step clinical inference.
  • A learned gate mixes the visual and symbolic signals before a BioGPT decoder writes the report.

On the small COVID-19 CT benchmark COV-CTR, the model is competitive with or better than prior systems on most metrics, and ablations show both retrieval and reasoning help. Attention maps and reasoning paths make the output easier to audit. However, retrieval quality is not measured quantitatively, and evaluation covers only one dataset.

Key points

  • Data and inputs: evaluation uses only COV-CTR (728 2D chest CT images, mostly COVID-19), and the knowledge source is PrimeKG.
  • Pipeline: ViT encoder → VSR retrieval, trained with a contrastive loss against a graph prototype → GAT subgraph encoding with graph-to-vision cross-attention → GRU multi-hop reasoning → gated fusion → BioGPT decoder.
  • BLEU-4 results: the model scores 0.625, compared with 0.535 for a visual-only baseline and 0.570 for ASGK. This is the second-highest BLEU-4 among the compared systems.
  • Other metrics and ablations: the model achieves the best BLEU-1, BLEU-2, METEOR and ROUGE-L. Removing either latent retrieval or the Multi-Hop Reasoning Module noticeably lowers scores.
  • Stated limitations:
    • Coverage is bounded by what PrimeKG contains.
    • A single retrieval prototype handles comorbidities poorly.
    • Volumetric 3D information is not used.
    • The reasoning module adds inference cost.
    • There are no quantitative metrics for reasoning paths; the authors propose graph edit distance for future work.

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

 Institutions

🎓University of ExeterExeter · United Kingdom12 works · 203 citations

 Fields, topics and concepts

📚 Shelves 7The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksComputer Vision and Pattern Recognition · 2,869 worksMedicine · 241 worksHealth Informatics · 184 works

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

Cite

Cite as Faezeh Safari, Hang Dong, Zeyu Fu and Aline Villavicencio. 2026. GraphRAG-Rad: Concept-Aware Radiology Report Generation via Latent Visual-Semantic Retrieval. Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 4: Student Research Workshop).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.eacl-srw.34

@inproceedings{safari2026graphrag,
  title = {{GraphRAG-Rad: Concept-Aware Radiology Report Generation via Latent Visual-Semantic Retrieval}},
  author = {Faezeh Safari, Hang Dong, Zeyu Fu and Aline Villavicencio},
  booktitle = {Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 4: Student Research Workshop)},
  year = {2026},
  doi = {10.18653/v1/2026.eacl-srw.34},
  url = {https://doi.org/10.18653/v1/2026.eacl-srw.34}
}

References

36 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Zaheer Babar, Twan van Laarhoven, and E. Marchiori. 2021. Encoder-decoder models for chest x-ray report generation perform no better than unconditioned baselines. PLoS ONE, 16. find
  2. Payal Chandak, Kexin Huang, and Marinka Zitnik. 2023. Building a knowledge graph to enable precision medicine. Scientific Data, 10(1):67.
  3. Weixing Chen, Yang Liu, Ce Wang, Jiarui Zhu, Guanbin Li, Cheng-Lin Liu, and Liang Lin. 2025. Cross-modal causal representation learning for radiology report generation. IEEE Transactions on Image Processing. find
  4. Zhihong Chen, Yan Song, Tsung-Hui Chang, and Xiang Wan. 2020. Generating radiology reports via memory-driven transformer. arXiv preprint arXiv:2010.16056.
  5. Marcella Cornia, Matteo Stefanini, Lorenzo Baraldi, and Rita Cucchiara. 2020. Meshed-memory transformer for image captioning. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10578–10587. find
  6. Peketi Divya, Yenduri Sravani, Chalavadi Vishnu, C Krishna Mohan, and Yen Wei Chen. 2024. Memory guided transformer with spatio-semantic visual extractor for medical report generation. IEEE Journal of Biomedical and Health Informatics, 28(5):3079– 3089. find
  7. Anna Fink, Johanna Nattenmüller, Stephan Rau, Alexander Rau, Hien Tran, Fabian Bamberg, Marco Reisert, Elmar Kotter, Thierno Diallo, and Maximilian F Russe. 2025. Retrieval-augmented generation improves precision and trust of a gpt-4 model for emergency radiology diagnosis and classification: a proo… find
  8. Yu Gu, Robert Tinn, Hao Cheng, Michael Lucas, Naoto Usuyama, Xiaodong Liu, Tristan Naumann, Jianfeng Gao, and Hoifung Poon. 2021. Domain-specific language model pretraining for biomedical natural language processing. ACM Transactions on Computing for Healthcare (HEALTH), 3(1):1–23.
  9. Navdeep Kaur and Ajay Mittal. 2022. Chexprune: sparse chest x-ray report generation model using multi-attention and one-shot global pruning. Journal of Ambient Intelligence and Humanized Computing, 14:7485 – 7497. find
  10. Jacob Devlin Ming-Wei Chang Kenton, Lee Kristina Toutanova, and 1 others. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of naacL-HLT, volume 1. Minneapolis, Minnesota.
  11. Mingjie Li, Rui Liu, Fuyu Wang, Xiaojun Chang, and Xiaodan Liang. 2023. Auxiliary signal-guided knowledge encoder-decoder for medical report generation. World Wide Web, 26(1):253–270. find
  12. Fenglin Liu, Xian Wu, Shen Ge, Wei Fan, and Yuexian Zou. 2021. Exploring and distilling posterior and prior knowledge for radiology report generation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 13753– 13762. find
  13. Renqian Luo, Liai Sun, Yingce Xia, Tao Qin, Sheng Zhang, Hoifung Poon, and Tie-Yan Liu. 2022. Biogpt: generative pre-trained transformer for biomedical text generation and mining. Briefings in bioinformatics, 23(6):bbac409. find
  14. Jong Hak Moon, Hyungyung Lee, Woncheol Shin, Young-Hak Kim, and Edward Choi. 2022. Multimodal understanding and generation for medical images and text via vision-language pre-training. IEEE Journal of Biomedical and Health Informatics, 26(12):6070–6080. find
  15. Peng Qi, Yuhao Zhang, Yuhui Zhang, Jason Bolton, and Christopher D Manning. 2020. Stanza: A python natural language processing toolkit for many human languages. arXiv preprint arXiv:2003.07082.
  16. Santhosh Raminedi, S. Shridevi, and Daehan Won. 2024. Multi-modal transformer architecture for medical image analysis and automated report generation. Scientific Reports, 14. find
  17. Prateek Singh and Sudhakar Singh. 2025. Chestx-transcribe: a multimodal transformer for automated radiology report generation from chest x-rays. Frontiers in Digital Health, 7. find
  18. Mehreen Sirshar, Muhammad Faheem Khalil Paracha, M. Akram, N. Alghamdi, S. Z. Y. Zaidi, and Tatheer Fatima. 2022. Attention based automated radiology report generation using cnn and lstm. PLoS ONE, 17. find
  19. Yun Tan, Chunzhi Li, Jiaohua Qin, Youyuan Xue, and Xuyu Xiang. 2024. Medical image description based on multimodal auxiliary signals and transformer. International Journal of Intelligent Systems, 2024(1):6680546. find
  20. Yuhao Tang and Fei Tao. 2025. From coarse to grain: automated medical report generation based on fine-grained semantic alignment and cross-modal enhancement. Cluster Computing, 28(10):636. find
  21. Yitian Tao, Liyan Ma, Jing Yu, and Han Zhang. 2024. Memory-based cross-modal semantic alignment network for radiology report generation. IEEE Journal of Biomedical and Health Informatics, 28(7):4145– 4156. find
  22. Steffanie S Weinreich, R Mangon, JJ Sikkens, ME En Teeuw, and MC Cornel. 2008. Orphanet: a european database for rare diseases. Nederlands tijdschrift voor geneeskunde, 152(9):518–519. find
  23. Xing Wu, Jingwen Li, Jianjia Wang, and Quan Qian. 2023. Multimodal contrastive learning for radiology report generation. Journal of Ambient Intelligence and Humanized Computing, 14(8):11185–11194. find
  24. Liming Xu, Yongheng Wang, Chunlin He, Quan Tang, Xianhua Zeng, and Jiancheng Lv. 2025. Deep disease label-guided graph convolutional network for medical report generation. ACM Transactions on Knowledge Discovery from Data, 19(5):1–23. find
  25. Sixing Yan, William K Cheung, Keith Chiu, Terence M Tong, Ka Chun Cheung, and Simon See. 2023. Attributed abnormality graph embedding for clinically accurate x-ray report generation. IEEE Transactions on Medical Imaging, 42(8):2211–2222. find
  26. Carl Yang, Hejie Cui, Jiaying Lu, Shiyu Wang, Ran Xu, Wenjing Ma, Yue Yu, Shaojun Yu, Xuan Kan, Chen Ling, and 1 others. 2023a. A review on knowledge graphs for healthcare: Resources, applications, and promises. arXiv preprint arXiv:2306.04802. — [arXiv:2306.04802](https://arxiv.org/abs/2306.04802) find
  27. Shuxin Yang, Xian Wu, Shen Ge, Zhuozhao Zheng, S Kevin Zhou, and Li Xiao. 2023b. Radiology report generation with a learned knowledge base and multi-modal alignment. Medical Image Analysis, 86:102798. find
  28. Xingyi Yang, Xuehai He, Jinyu Zhao, Yichen Zhang, Shanghang Zhang, and Pengtao Xie. 2020. Covid-ct-dataset: a ct scan dataset about covid-19. arXiv preprint arXiv:2003.13865. — [arXiv:2003.13865](https://arxiv.org/abs/2003.13865) find
  29. Yan Yang, Xiaoxing You, Ke Zhang, Zhenqi Fu, Xianyun Wang, Jiajun Ding, Jiamei Sun, Zhou Yu, Qingming Huang, Weidong Han, and 1 others. 2025. Spatio-temporal and retrieval-augmented modelling for chest x-ray report generation. IEEE Transactions on Medical Imaging. find
  30. Ke Zhang, Hanliang Jiang, Jian Zhang, Qingming Huang, Jianping Fan, Jun Yu, and Weidong Han. 2023. Semi-supervised medical report generation via graph-guided hybrid feature consistency. IEEE Transactions on Multimedia, 26:904–915. find
  31. Ke Zhang, Yan Yang, Jun Yu, Jianping Fan, Hanliang Jiang, Qingming Huang, and Weidong Han. 2024a. Attribute prototype-guided iterative scene graph for explainable radiology report generation. IEEE Transactions on Medical Imaging. find
  32. Wenfeng Zhang, Baoning Cai, Jianming Hu, Qibing Qin, and Kezhen Xie. 2024b. Visual-textual cross-modal interaction network for radiology report generation. IEEE Signal Processing Letters, 31:984–988. find
  33. Yuhao Zhang, Yuhui Zhang, Peng Qi, Christopher D Manning, and Curtis P Langlotz. 2021. Biomedical and clinical english model packages for the stanza python nlp library. Journal of the American Medical Informatics Association, 28(9):1892–1899. find
  34. Ziqi Zhang and Ailian Jiang. 2024. Interactive dual-stream contrastive learning for radiology report generation. Journal of Biomedical Informatics, 157:104718. find
  35. Junting Zhao, Yang Zhou, Zhihao Chen, Huazhu Fu, and Liang Wan. 2024. Topicwise separable sentence retrieval for medical report generation. IEEE Transactions on Medical Imaging. find
  36. Hong-Yu Zhou, Xiaoyu Chen, Yinghao Zhang, Ruibang Luo, Liansheng Wang, and Yizhou Yu. 2022. Generalized radiograph representation learning via cross-supervision between images and free-text radiology reports. Nature Machine Intelligence, 4(1):32–40. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/safari-2026-graphrag-rad-concept-aware-radiology/