Mova0.5 pre-release⌕
☰
home / docs / papers / Modern Hebrew / Hierarchical Section Matching Prediction (HSMP) BERT for Fine-Grained Extraction of Structured Data from Hebrew Free-Text Radiology Reports in Crohn's Disease

Hierarchical Section Matching Prediction (HSMP) BERT for Fine-Grained Extraction of Structured Data from Hebrew Free-Text Radiology Reports in Crohn's Disease

What it is about

The authors adapt a Hebrew RoBERTa model to 9,683 Crohn's disease radiology reports, using a masked-language objective and a second objective that predicts whether a Findings section and an Impression section belong to the same report. The model then answers prompts of the form "There is {finding} in the {organ}" for each organ–finding pair. At inference it asks top-down questions (scan normal? organ abnormal? finding present?), and a negative answer skips the questions below it. On 512 manually annotated reports, with 24 labels that have more than 15 positives, the mean F1 is 0.83. The zero-shot version scores 0.49 and standard fine-tuning 0.30. The hierarchical scheme runs 5.1 times faster than the flat one in total time, and the paper uses it to report co-occurrence patterns over the whole corpus.

Ideas

For Mova

I1 belongs to world (the order in which an extractor asks its questions) and to prag (the frame rules). I2 is a test for world. I3 is the reporting standard for every extraction test in prag.

Cite

The paper itself, not this card. BibTeX from the library's record of the work. arXiv 2509.04519

@misc{turner2025hierarchical,
  title = {{Hierarchical Section Matching Prediction (HSMP) BERT for Fine-Grained Extraction of Structured Data from Hebrew Free-Text Radiology Reports in Crohn's Disease}},
  author = {Turner;, Zvi Badash; Hadas Ben-Atya; Naama Gavrielov; Liam Hazan; Gili Focht; Ruth Cytter-Kuint; Talar Hagopian; Dan and Freiman, Moti},
  year = {2025},
  eprint = {2509.04519},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2509.04519}
}

References

34 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. P. Reschke, C. Booz, L. D. Gruenewald, V. Koch, E. Höhne, A. Gökduman, K. Eichler, J. Schlüchtermann, T. J. Vogl, and J. Gotta, “Rethinking radiology reports: A survey of referring physicians’ perspectives,” European Journal of Radiology, vol. 187, 6 2025. find
  2. T. Jorg, M. C. Halfmann, N. Rölz, R. Mager, D. P. dos Santos, C. Düber, P. Mildenberger, and L. Müller, “Structured reporting in radiology enables epidemiological analysis through data mining: urolithiasis as a use case,” Abdominal Radiology, vol. 48, pp. 3520–3529, 11 2023. find
  3. A. Mozayan, A. R. Fabbri, M. Maneevese, I. Tocino, and S. Chheang, “Practical guide to natural language processing for radiology,” Radiographics, vol. 41, pp. 1446–1453, 9 2021. find
  4. S. Nowak, D. Biesner, Y. C. Layer, M. Theis, H. Schneider, W. Block, B. Wulff, U. I. Atten-berger, R. Sifa, and A. M. Sprinkart, “Transformer-based structuring of free-text radiology report databases,” European Radiology, vol. 33, pp. 4228–4236, 6 2023. find
  5. M. M. Graf, K. K. Bressem, and L. C. Adams, “Transformation of free-text radiology reports into structured data,” Radiologie, vol. 65, pp. 249–256, 4 2025. find
  6. P. Woźnicki, C. Laqua, I. Fiku, A. Hekalo, D. Truhn, S. Engelhardt, J. Kather, S. Foersch, T. A. D’Antonoli, D. P. dos Santos, B. Baeßler, and F. C. Laqua, “Automatic structuring of radiology reports with on-premise open-source large language models,” European Radiology, vol. 35, pp. 2018–2029, 4 2025. find
  7. D. R. Schlegel, K. Gordon, C. Gaudioso, and M. Peleg, “Clinical tractor: A framework for automatic natural language understanding of clinical practice guidelines,” AMIA Annual Symposium Proceedings, vol. 2019, pp. 784–793, 3 2020. find
  8. D. Reichenpfader, H. Müller, and K. Denecke, “A scoping review of large language model based approaches for information extraction from radiology reports,” npj Digital Medicine 2024 7:1, vol. 7, pp. 1–12, 8 2024. find
  9. B. L. Guellec, A. Lefèvre, C. Geay, L. Shorten, C. Bruge, L. Hacein-Bey, P. Amouyel, J. P. Pruvo, G. Kuchcinski, and A. Hamroun, “Performance of an open-source large language model in extracting information from free-text radiology reports,” Radiology: Artificial Intelligence, vol. 6, 7 2024. find
  10. S. Liu, A. B. McCoy, and A. Wright, “Improving large language model applications in biomedicine with retrieval-augmented generation: a systematic review, meta-analysis, and clinical development guidelines,” Journal of the American Medical Informatics Association, vol. 32, pp. 605–615, 4 2025. find
  11. S. T. Arasteh, M. Lotfinia, K. Bressem, R. Siepmann, L. Adams, D. Ferber, C. Kuhl, J. N. Kather, S. Nebelung, and D. Truhn, “Radiorag: Online retrieval-augmented generation for radiology question answering,” Radiology. Artificial intelligence, vol. 7, 7 2025. find
  12. M. G. Linguraru, S. Bakas, M. Aboian, P. D. Chang, A. E. Flanders, J. Kalpathy-Cramer, F. C. Kitamura, M. P. Lungren, J. Mongan, L. M. Prevedello, R. M. Summers, C. C. Wu, M. Adewole, and C. E. Kahn, “Clinical, cultural, computational, and regulatory considerations to deploy ai in radiology: Perspectives of rsna and miccai experts,” Radiology: Artificial Intelligence, vol. 6, 7 2024. find
  13. Y. Artsi, V. Sorin, and E. Klang, “Challenges in using large language models for simplifying radiology reports,” Journal of the American College of Radiology, vol. 22, p. 405, 4 2025. find
  14. F. Dennstädt, J. Hastings, P. M. Putora, M. Schmerder, and N. Cihoric, “Implementing large language models in healthcare while balancing control, collaboration, costs and security,” npj Digital Medicine, vol. 8, pp. 1–4, 12 2025. find
  15. F. Busch, L. Hoffmann, C. Rueger, E. H. van Dijk, R. Kader, E. Ortiz-Prado, M. R. Makowski, L. Saba, M. Hadamitzky, J. N. Kather, D. Truhn, R. Cuocolo, L. C. Adams, and K. K. Bressem, “Current applications and challenges in large language models for patient care: a systematic review,” Communications Medicine 2025 5:1, vol. 5, pp. 1–13, 1 2025. find
  16. J. C. L. Ong, S. Y.-H. Chang, W. William, A. J. Butte, N. H. Shah, L. S. T. Chew, N. Liu, F. Doshi-Velez, W. Lu, J. Savulescu, and D. S. W. Ting, “Medical ethics of large language models in medicine,” NEJM AI, vol. 1, p. 14, 6 2024. find
  17. T. A. D’Antonoli, A. S. Tejani, B. Khosravi, C. Bluethgen, F. Busch, K. K. Bressem, L. C. Adams, M. Moassefi, S. Faghani, and J. W. Gichoya, “Cybersecurity threats and mitigation strategies for large language models in health care,” Radiology. Artificial intelligence, vol. 7, 7 2025. find
  18. E. Malul, O. Perets, Z. Mor, Y. Kassel, and E. Sulem, “Lchaim - investigating long context reasoning in hebrew,” in Findings of the Association for Computational Linguistics: ACL 2025 (W. Che, J. Nabende, E. Shutova, and M. T. Pilehvar, eds.), pp. 7928–7939, Association for Computational Linguistics, 7 2025. find our card
  19. J. Sha, M. Zhu, C. Feng, and Y. Shang, “Veef-multi-llm: Effective vocabulary expansion and parameter efficient finetuning towards multilingual large language models,” in Proceedings of the 31st International Conference on Computational Linguistics (O. Rambow, L. Wanner, M. Apidianaki, H. Al-Khalifa, B. D. Eugenio, and S. Schockaert, eds.), pp. 7963–7981, Association for Computational Linguistics, 1 2025. find
  20. H. Y. Li, J. F. Fu, and A. Python, “Implementing large language models in health care: Clinician-focused review with interactive guideline,” Journal of medical Internet research, vol. 27, p. e71916, 7 2025. find
  21. E. Avnat, M. Levy, D. Herstain, E. Yanko, D. B. Joya, M. T. Katz, D. Eshel, S. Laros, Y. Dagan, S. Barami, J. Mermelstein, S. Ovadia, N. Shomron, V. Shalev, and R. E. E. Abdulnour, “Performance of large language models in numerical versus semantic medical knowledge: Cross-sectional benchmarking study on evidence-based questions and answers,” Journal of Medical Internet Research, vol. 27, 2025. find
  22. D. H. Bruining, E. M. Zimmermann, E. V. Loftus Jr, W. J. Sandborn, C. G. Sauer, S. A. Strong, and S. of Abdominal Radiology Crohn’s Disease-Focused Panel, “Consensus recommendations for evaluation, interpretation, and utilization of computed tomography and magnetic resonance enterography in patients with small bowel crohn’s disease,” Radiology, vol. 286, no. 3, pp. 776–799, 2018. find
  23. T. Jiang, J. Jiao, S. Huang, Z. Zhang, D. Wang, F. Zhuang, F. Wei, H. Huang, D. Deng, and Q. Zhang, “Promptbert: Improving bert sentence embeddings with prompts,” in Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing (Y. Goldberg, Z. Kozareva, and Y. Zhang, eds.), pp. 8826–8837, Association for Computational Linguistics, 12 2022. find
  24. Y. Zhuang, J. Zhang, X. Li, C. Liu, Y. Yu, W. Dong, and K. He, “Autonomous international classification of diseases coding using pretrained language models and advanced prompt learning techniques: Evaluation of an automated analysis system using medical text,” JMIR Medical Informatics, vol. 13, 2025. find
  25. D. Reichenpfader, J. Knupp, S. U. von Däniken, R. Gaio, F. Dennstädt, G. M. Cereghetti, A. Sander, H. Hiltbrunner, K. Nairz, and K. Denecke, “Enhancing bidirectional encoder representations from transformers (bert) with frame semantics to extract clinically relevant information from german mammography reports: Algorithm development and validation,” Journal of Medical Internet Research, vol. 27, 2025. find
  26. L. Hazan, N. Gavrielov, R. Reichart, T. Hagopian, M.-L. Greer, R. Cytter-Kuint, G. Focht, D. Turner, and M. Freiman, “Leveraging prompt-learning for structured information extraction from crohn’s disease radiology reports in a low-resource language,” in Proceedings of the 6th Clinical Natural Language Processing Workshop (T. Naumann, A. B. Abacha, S. Bethard, K. Roberts, and D. Bitterman, eds.), pp. 301–309, Association for Computational Linguistics, 6 2024. find
  27. M. Y. Friedman, M. Leventer-Roberts, J. Rosenblum, N. Zigman, I. Goren, V. Mourad, N. Lederman, N. Cohen, E. Matz, D. Z. Dushnitzky, N. Borovsky, M. B. Hoshen, G. Focht, M. Avitzour, Y. Shachar, Y. Chowers, R. Eliakim, S. Ben-Horin, S. Odes, D. Schwartz, I. Dotan, E. Israeli, Z. Levi, E. I. Benchimol, R. D. Balicer, and D. Turner, “Development and validation of novel algorithms to identify patients with inflammatory bowel diseases in israel: an epi-iirn group study,” Clinical Epidemiology, vol. 10, p. 671, 6 2018. find
  28. V. Shalumov and H. Haskey, “Hero: Roberta and longformer hebrew language models,” arXiv preprint arXiv:2304.11077, 2023. ↗ our card
  29. S. Gururangan, A. Marasović, S. Swayamdipta, K. Lo, I. Beltagy, D. Downey, and N. A. Smith, “Don’t stop pretraining: Adapt language models to domains and tasks,” in Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (D. Jurafsky, J. Chai, N. Schluter, and J. Tetreault, eds.), pp. 8342–8360, Association for Computational Linguistics, 7 2020. find
  30. J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “Bert: Pre-training of deep bidirectional transformers for language understanding,” in Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) (J. Burstein, C. Doran, and T. Solorio, eds.), pp. 4171–4186, Association for Computational Linguistics, 6 2019. find
  31. J. Lee, W. Yoon, S. Kim, D. Kim, S. Kim, C. H. So, and J. Kang, “Biobert: A pre-trained biomedical language representation model for biomedical text mining,” Bioinformatics, vol. 36, pp. 1234–1240, 2 2020. find
  32. K. Huang, J. Altosaar, and R. Ranganath, “Clinicalbert: Modeling clinical notes and predicting hospital readmission,” arXiv preprint arXiv:1904.05342, 2019. ↗
  33. K. S. Kalyan, “A survey of gpt-3 family large language models including chatgpt and gpt-4,” Natural Language Processing Journal, vol. 6, p. 100048, 3 2024. find
  34. G. Al-Kateb, E. Cengiz, and M. Gök, “Biogpt: A generative transformer-based framework for personalized genomic medicine and rare disease diagnosis,” Mesopotamian Journal of Artificial Intelligence in Healthcare, vol. 2025, pp. 154–164, 7 2025. Supporting information Table S1: Label prevalence for each of the selected labels (with at least 15 positive examples among the IBD cases). Note: Some labels are near-perfectly balanced (e.g., “Ileum Wall Thickness”), but most suffer from extreme label imbalance with < 5% positive examples in the annotated dataset.. Label Ileum Wall Thickness Ileum Inflammation Ileum Wall Enhancement Ileum Stenosis Ileum Pre Stenotic Dilatation Colon Inflammation Ileum DWI Signal Ileum Comb sign Sigmoid Inflammation Colon Wall Thickness Ileum Mesenteric Edema Sigmoid Wall Thickness Colon Wall Enhancement Rectum Inflammation Sigmoid Wall Enhancement Ileum Fistula Cecum Inflammation Rectum Wall Thickness Cecum Wall Thickness Cecum Wall Enhancement Rectum Wall Enhancement Ileum Reduced Motility Sigmoid Comb sign Sigmoid Mesenteric Edema Prevalence 45.2% (215/476) 41.8% (199/476) 34.7% (165/476) 24.2% (115/476) 17.9% (85/476) 15.5% (74/476) 14.7% (70/476) 14.5% (69/476) 13.7% (65/476) 13.4% (64/476) 12.8% (61/476) 11.6% (55/476) 8.8% (42/476) 8.0% (38/476) 7.4% (35/476) 7.1% (34/476) 7.1% (34/476) 6.7% (32/476) 6.7% (32/476) 4.8% (23/476) 4.4% (21/476) 3.4% (16/476) 3.4% (16/476) 3.2% (15/476) Table S2: Performance metrics of the SMP-BERT model across the 24 organ–finding combinations. NPV = negative predictive value, PPV = positive predictive value. Label Accuracy F1 Score Cecum Inflammation Cecum Wall Enhancement Cecum Wall Thickness Colon Inflammation Colon Wall Enhancement Colon Wall Thickness Ileum Comb sign Ileum DWI Signal Ileum Fistula Ileum Inflammation Ileum Mesenteric Edema Ileum Pre Stenotic Dilatation Ileum Reduced Motility Ileum Stenosis Ileum Wall Enhancement Ileum Wall Thickness Rectum Inflammation Rectum Wall Enhancement Rectum Wall Thickness Sigmoid Comb sign Sigmoid Inflammation Sigmoid Mesenteric Edema Sigmoid Wall Enhancement Sigmoid Wall Thickness 0.96 0.97 0.96 0.94 0.94 0.93 0.91 0.97 0.98 0.87 0.84 0.94 0.98 0.95 0.94 0.92 0.94 0.96 0.97 0.97 0.87 0.97 0.94 0.89 Cohen’s κ Balanced Accuracy NPV PPV 0.84 0.83 0.84 0.88 0.79 0.85 0.80 0.93 0.93 0.87 0.65 0.91 0.78 0.93 0.93 0.92 0.83 0.80 0.88 0.63 0.75 0.74 0.77 0.74 0.67 0.65 0.68 0.77 0.58 0.69 0.60 0.87 0.87 0.73 0.30 0.81 0.56 0.87 0.86 0.85 0.66 0.61 0.77 0.27 0.51 0.49 0.54 0.48 0.85 0.78 0.84 0.90 0.77 0.83 0.77 0.91 0.99 0.88 0.65 0.91 0.70 0.95 0.94 0.93 0.89 0.84 0.94 0.60 0.79 0.70 0.76 0.75 0.98 0.98 0.98 0.97 0.96 0.95 0.93 0.97 1.00 0.94 0.91 0.97 0.98 0.98 0.98 0.97 0.99 0.99 0.99 0.97 0.95 0.98 0.97 0.94 0.67 0.80 0.70 0.78 0.67 0.79 0.76 0.95 0.79 0.79 0.38 0.83 1.00 0.86 0.87 0.87 0.59 0.66 0.69 0.52 0.52 0.67 0.60 0.53 Table S3: Performance metrics of the SMP/BERT model when treated as a zero-shot model (no SMP fine-tuning), across the same 24 organ–finding combinations. NPV = negative predictive value, PPV = positive predictive value. Label Accuracy F1 Score Cecum Inflammation Cecum Wall Enhancement Cecum Wall Thickness Colon Inflammation Colon Wall Enhancement Colon Wall Thickness Ileum Comb sign Ileum DWI Signal Ileum Fistula Ileum Inflammation Ileum Mesenteric Edema Ileum Pre Stenotic Dilatation Ileum Reduced Motility Ileum Stenosis Ileum Wall Enhancement Ileum Wall Thickness Rectum Inflammation Rectum Wall Enhancement Rectum Wall Thickness Sigmoid Comb sign Sigmoid Inflammation Sigmoid Mesenteric Edema Sigmoid Wall Enhancement Sigmoid Wall Thickness 0.52 0.61 0.66 0.68 0.68 0.76 0.60 0.49 0.72 0.53 0.63 0.67 0.77 0.59 0.61 0.62 0.59 0.65 0.77 0.61 0.65 0.71 0.65 0.79 Cohen’s κ Balanced Accuracy NPV PPV 0.39 0.42 0.46 0.57 0.52 0.61 0.49 0.44 0.52 0.51 0.48 0.58 0.44 0.51 0.57 0.60 0.38 0.43 0.46 0.42 0.51 0.45 0.47 0.57 -0.01 0.01 0.03 0.18 0.12 0.23 0.06 0.06 0.12 0.02 0.03 0.19 -0.06 0.03 0.15 0.21 -0.11 -0.01 -0.05 0.03 0.07 0.02 0.04 0.15 0.49 0.53 0.54 0.63 0.63 0.65 0.55 0.56 0.64 0.51 0.53 0.62 0.40 0.52 0.58 0.61 0.36 0.48 0.46 0.61 0.56 0.56 0.56 0.59 0.93 0.96 0.94 0.90 0.94 0.91 0.88 0.89 0.96 0.59 0.88 0.87 0.96 0.76 0.71 0.64 0.89 0.95 0.93 0.98 0.93 0.97 0.94 0.91 0.07 0.05 0.08 0.26 0.15 0.29 0.18 0.17 0.13 0.43 0.15 0.29 0.00 0.27 0.44 0.59 0.02 0.04 0.03 0.05 0.17 0.04 0.09 0.22 Table S4: Performance metrics of a standard fine-tune (SFT) BERT model—identical to the HSMP-BERT backbone—across the same 24 organ–finding combinations. NPV = negative predictive value, PPV = positive predictive value. Label Accuracy F1 Score Cecum Inflammation Cecum Wall Enhancement Cecum Wall Thickness Colon Inflammation Colon Wall Enhancement Colon Wall Thickness Ileum Comb sign Ileum DWI Signal Ileum Fistula Ileum Inflammation Ileum Mesenteric Edema Ileum Pre Stenotic Dilatation Ileum Reduced Motility Ileum Stenosis Ileum Wall Enhancement Ileum Wall Thickness Rectum Inflammation Rectum Wall Enhancement Rectum Wall Thickness Sigmoid Comb sign Sigmoid Inflammation Sigmoid Mesenteric Edema Sigmoid Wall Enhancement Sigmoid Wall Thickness 0.93 0.96 0.94 0.89 0.92 0.89 0.87 0.92 0.93 0.88 0.87 0.90 0.97 0.91 0.91 0.91 0.91 0.96 0.93 0.97 0.90 0.97 0.93 0.89 Cohen’s κ Balanced Accuracy NPV PPV 0.00 0.00 0.00 0.53 0.25 0.45 0.41 0.67 0.15 0.87 0.09 0.70 0.00 0.82 0.87 0.90 0.00 0.00 0.00 0.00 0.38 0.00 0.00 0.10 0.00 0.00 0.00 0.47 0.23 0.40 0.35 0.63 0.14 0.76 0.06 0.64 0.00 0.76 0.80 0.81 -0.02 0.00 -0.01 0.00 0.35 0.00 0.00 0.08 0.50 0.50 0.50 0.69 0.57 0.65 0.64 0.76 0.54 0.89 0.52 0.80 0.50 0.88 0.91 0.91 0.49 0.50 0.50 0.50 0.62 0.50 0.50 0.52 0.93 0.96 0.94 0.90 0.92 0.90 0.89 0.92 0.94 0.99 0.88 0.92 0.97 0.94 0.97 0.99 0.92 0.96 0.94 0.97 0.90 0.97 0.93 0.89 0.00 0.00 0.00 0.77 1.00 0.78 0.64 0.92 0.50 0.78 0.33 0.76 0.00 0.82 0.81 0.83 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.50 Figure S1: Each cell represents the pairwise Pearson correlation coefficient between the presence of two findings across patients. Rows and columns are ordered by anatomical location, from proximal (jejunum) to distal (rectum), facilitating interpretation of spatial anatomical patterns. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/langstream-2025-badash-hierarchical-section-matching-prediction-hsmp/