Elamur Research⌕
☰
home / papers / 2026 / Role-Adapted Clinical Report Generation for Ultrasound Measurements in Low-Resource Settings

Role-Adapted Clinical Report Generation for Ultrasound Measurements in Low-Resource Settings

Paper · 2026 · Ayoub Nainia, Tanya Akumu, Noussair Lazrak et al.

Summary

In many low- and middle-income countries, community health workers, nurses and midwives often take or receive fetal ultrasound measurements but lack the training to judge what they mean clinically. The authors build an offline system that turns fetal biometry values (HC, BPD, AC, FL, EFW) into clinical reports matched to the reader's role. It covers six healthcare worker roles grouped into three expertise levels.

The system has two separate paths:

  • Quantitative path: growth percentiles come from the published INTERGROWTH-21st equations, with Hadlock-based EFW. Fixed percentile cut-offs then produce red, yellow or green alerts, so the safety-critical urgency signal never depends on LLM output.
  • Narrative path: a locally run LLM writes the report text, grounded by RAG over a knowledge base drawn from the WHO Manual of Diagnostic Ultrasound. Role-specific prompt instructions set the vocabulary and level of detail.

Evaluation shows near-exact percentile computation, fully correct alert classification, readability that differs by role, and high factual consistency. The design matters because LLMs can produce confident errors, and in settings without specialist review, no one may catch them.

Key points

  • Knowledge base and retrieval: 171 entries sourced from Volume 2 of the WHO Manual of Diagnostic Ultrasound, encoded with paraphrase-multilingual-MiniLM-L12-v2 and stored in ChromaDB. Machine-translated Swahili entries serve as a proof of multilingual extensibility.
  • Generation backend: Llama 3 8B served through Ollama. The architecture does not depend on a particular LLM and runs without internet access.
  • Alert rules:
    • Red: any measurement below the 3rd or above the 97th percentile.
    • Yellow: any measurement in the 3rd–10th or 90th–97th band.
    • Green: all measurements between the 10th and 90th percentiles.
    • The worst flag across all measurements sets the overall severity.
  • Accuracy results: percentile computation is accurate to under a millimetre, and decision support classification is 100% correct.
  • Report quality results: Flesch-Kincaid grade is 8.8 for community health workers versus 11–13 for clinical roles. Factual consistency is 98% across 42 reports covering seven clinical scenarios.

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

🎓Sorbonne UniversitéParis · France32 works · 7,270 citations🎓Universitat de BarcelonaBarcelona · Spain12 works · 1,338 citations

Also on the byline: Institut de Systématique, Évolution, Biodiversité.

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Medicine · 241 worksHealth Informatics · 184 worksHealth Professions · 69 worksHealth Information Management · 50 worksPublic Health, Environmental and Occupational Health · 10 works

🧭 Topics 1Research subjects and languages the work is about.Low-resource languages · 267 works

Cite

Cite as Ayoub Nainia, Tanya Akumu, Noussair Lazrak and Karim Lekadir. 2026. Role-Adapted Clinical Report Generation for Ultrasound Measurements in Low-Resource Settings. Proceedings of the 8th Workshop on Clinical Natural Language Processing (Clinical NLP) @ LREC 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/28pa52zsmb39

@inproceedings{nainia2026role,
  title = {{Role-Adapted Clinical Report Generation for Ultrasound Measurements in Low-Resource Settings}},
  author = {Ayoub Nainia, Tanya Akumu, Noussair Lazrak and Karim Lekadir},
  booktitle = {Proceedings of the 8th Workshop on Clinical Natural Language Processing (Clinical NLP) @ LREC 2026},
  year = {2026},
  doi = {10.63317/28pa52zsmb39},
  url = {https://doi.org/10.63317/28pa52zsmb39}
}

References

21 references, 9 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Zhihong Chen, Yan Song, Tsung-Hui Chang, and Xiang 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. Association for Computational Linguistics.
  2. Tomasz Ciecierski-Holmes, Raj Singh, Matthew Axt, Stefan Axt, and Samia Brenner. 2022. Artificial intelligence for strengthening healthcare systems in low- and middle-income countries: a systematic scoping review. npj Digital Medicine, 5:162. find
  3. Sabrina Della Ripa, Christopher Oduor, Anthony Wanyoro, Peter Nguhiu, Elizabeth Rakers, and Cynthia K. Stanton. 2025. AI-enabled obstetric point-of-care ultrasound as an emerging technology in low- and middle-income countries: provider and health system perspectives. BMC Pregnancy and Childbirth, 25… find
  4. Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, et al. 2024. The Llama 3 herd of models.
  5. Frank P. Hadlock, R.B. Harrist, Ralph S. Sharman, Russell L. Deter, and Seung K. Park. 1985. Estimation of fetal weight with the use of head, body, and femur measurements—a prospective study. American Journal of Obstetrics and Gynecology, 151(3):333–337. find
  6. Katharina Jeblick, Balthasar Schachtner, Jakob Dexl, Andreas Mittermeier, Anna Theresa Stüber, Johanna Topalis, Tobias Weber, Philipp Wesp, Bastian Oliver Sabel, Jens Ricke, and Michael Ingrisch. 2024. ChatGPT makes medicine easy to swallow: an exploratory case study on simplified radiology reports.… find
  7. Alistair EW Johnson, Tom J Pollard, Lu Shen, Li-wei H Lehman, Mengling Feng, Mohammad Ghassemi, Benjamin Moody, Peter Szolovits, Leo Anthony Celi, and Roger G Mark. 2016. MIMIC-III, a freely accessible critical care database. Scientific Data, 3:160035.
  8. Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, and Douwe Kiela. 2020. Retrieval-augmented generation for knowledge-intensive NLP tasks. Advances in Neural Information Process…
  9. Yuan Li, Xiaodan Liang, Zhiting Hu, and Eric P Xing. 2018. Hybrid retrieval-generation reinforced agent for medical image report generation. In Advances in Neural Information Processing Systems, volume 31. find
  10. Don Nutbeam. 2008. The evolving concept of health literacy. Social Science & Medicine, 67(12):2072–2078. find
  11. Ankit Pal, Logesh Kumar Umapathi, and Malaikannan Sankarasubbu. 2023. Med-HALT: Medical domain hallucination test for large language models. In Proceedings of the 27th Conference on Computational Natural Language Learning (CoNLL), pages 314–334. Association for Computational Linguistics.
  12. Aris T Papageorghiou, Eric O Ohuma, Douglas G Altman, Tullia Todros, Leila Cheikh Ismail, Ann Lambert, Yasmin A Jaffer, Enrico Bertino, Michael G Gravett, Manorama Purwar, et al. 2014. International standards for fetal growth based on serial ultrasound measurements: the Fetal Growth Longitudinal Stu… find
  13. Nils Reimers and Iryna Gurevych. 2019. Sentence-BERT: Sentence embeddings using Siamese BERT-networks. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 3982–3992.…
  14. Karan Singhal, Shekoofeh Azizi, Tao Tu, S. Sara Mahdavi, Jason Wei, Hyung Won Chung, Nathan Scales, Ajay Tanwani, Heather Cole-Lewis, Stephen Pfohl, Perry Payne, Martin Seneviratne, Paul Gamble, Chris Kelly, Abubakr Babiker, Nathanael Schärli, Aakanksha Chowdhery, Philip Mansfield, Dina Demner-Fushm… — Large Language Models Encode Clinical Knowledge
  15. Isabella Catharina Wiest, Dyke Ferber, Jiefu Zhu, Marko van Treeck, Sonja K. Meyer, Radhika Juglan, Zunamys I. Carrero, Daniel Paech, Jens Kleesiek, Matthias P. Ebert, Daniel Truhn, and Jakob Nikolas Kather. 2024. Privacy-preserving large language models for structured medical information retrieval.… find
  16. World Health Organization. 2013. Manual of Diagnostic Ultrasound, 2nd edition, volume 2. World Health Organization, Geneva. find
  17. Guangzhi Xiong, Qiao Jin, Zhiyong Lu, and Aidong Zhang. 2024. Benchmarking retrieval-augmented generation for medicine. In Findings of the Association for Computational Linguistics: ACL 2024, pages 6233–6251. Association for Computational Linguistics.
  18. Aidin Spina, Saman Andalib, Daniel Flores, Rishi Vermani, Faris F Halaseh, and Ariana M Nelson. 2024. Evaluation of generative language models in personalizing medical information: Instrument validation study. JMIR AI, 3:e54371. find
  19. Julien Stirnemann, Laurent J Salomon, and Aris T Papageorghiou. 2020. INTERGROWTH-21st standards for Hadlock’s estimation of fetal weight. Ultrasound in Obstetrics & Gynecology, 56(6):946–948. find
  20. Arun James Thirunavukarasu, Daniel Shu Wei Ting, Kabilan Elangovan, Laura Gutierrez, Ting Fang Tan, and Daniel Shu Wei Ting. 2023. Large language models in medicine. Nature Medicine, 29:1930–1940. find
  21. Dave Van Veen, Cara Van Uden, Louis Blankemeier, Jean-Benoit Delbrouck, Asad Aali, Christian Bluethgen, Anuj Pareek, Malgorzata Polacin, Eduardo Pontes Reis, Anna Seehofnerová, Nidhi Rohatgi, Poonam Hosamani, William Collins, Neera Ahuja, Curtis P. Langlotz, Jason Hom, Sergios Gatidis, John Pauly, a… — Adapted Large Language Models Can Outperform Medical Experts in Clinical Text Summarization

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/nainia-2026-role-adapted-clinical-report-generation/