癫痫临床文本信息的结构化标注与抽取研究^*

Structured Annotation and Information Extraction of Epilepsy Clinical Texts

摘要

目的 面向癫痫专科临床及科研需求，构建一套细粒度、高覆盖的中文临床文本命名实体标注体系，并验证其在命名实体识别（named entity recognition，NER）任务中的有效性。方法 从疾病、病程时间、临床表现、医疗检查、非药物治疗、用药及影响因素等7大维度设计3级标注体系，共制定25类标签，明确标签边界与特殊表达处理规则。以四川大学华西医院2009～2023年癫痫住院患者脱敏病历为数据来源，由3名具有癫痫临床背景的标注员完成804例高质量语料标注，采用双人复标、专家仲裁与实体级标注者一致性（inter-annotator agreement，IAA）评估保障标注质量。以五种中文医学预训练语言模型（Base-BERT、chinese-bert、chinese-Roberta、MC-BERT、MedBERT）分别结合BiLSTM-CRF与GlobalPointer两种序列标注框架，共10种组合方案进行NER性能验证，并在外部文献病例集上开展跨域泛化评估。结果 最终语料库涵盖25类癫痫相关实体，标注实例804条，实体总数28 400个；三位标注员IAA为0.86～0.88，标注一致性达到计算语言学通行标准。NER验证中，GlobalPointer框架表现优于BiLSTM-CRF，整体Micro-F1为0.906、Macro-F1为0.760；高频核心实体（如发作症状、药物名称、时间信息）F1值均超过0.90；跨域文献病例验证中高频标签F1值维持在0.80以上，低频标签（如fac-inc-amb、trt、dru-adv）F1值为0.31～0.57，主要受样本量有限与表达多样性影响。结论 本研究构建的癫痫专科中文临床标注体系覆盖范围广、粒度细、一致性高，可有效支持NER模型的训练与评估，为癫痫病历结构化分析及下游智能诊疗工具的开发提供了可复用的语料基础。

Abstract

Objective To develop a fine-grained, highly comprehensive Chinese clinical texts named entity annotation schema tailored to the needs of epilepsy specialty clinical practice and research, and to validate its effectiveness in named entity recognition (NER) tasks. Methods A three-level annotation schema covering 25 entity types was designed across seven major dimensions, including disease, disease course timeline, clinical manifestations, medical examinations, non-pharmacological treatments, medication, and influencing factors, with explicit label boundary definitions and rules for handling special expressions. De-identified inpatient records of epilepsy patients admitted to West China Hospital, Sichuan University from 2009 to 2023 served as the data source. Three annotators with epilepsy clinical backgrounds completed high-quality annotation of 804 cases, with annotation quality ensured through double annotation, expert arbitration, and entity-level inter-annotator agreement (IAA) evaluation. NER performance was validated using 10 model combinations comprising five Chinese medical pre-trained language models (Base-BERT, chinese-bert, chinese-Roberta, MC-BERT, MedBERT) paired with two sequence labeling frameworks (BiLSTM-CRF and GlobalPointer), with additional cross-domain generalization evaluation on an external case dataset established on the basis of published literature. Results The final corpus contains 25 categories of epilepsy-related entities, 804 annotated cases, and a total of 28 400 entities. The IAA among the three annotators ranged from 0.86 to 0.88, indicating annotation consistency that met the accepted standards in computational linguistics. In NER validation, the GlobalPointer framework outperformed BiLSTM-CRF, achieving an overall Micro-F1 score of 0.906 and Macro-F1 score of 0.760. High-frequency core entities (e.g., seizure symptoms, drug names, and temporal information) all yielded F1 scores exceeding 0.90. In cross-domain validation on literature-based cases, high-frequency entity F1 scores remained above 0.80, while low-frequency entities (e.g., factors with incomplete/ambiguous induction [fac-inc-amb], treatment information [trt], and adverse drug reactions [dru-adv]) achieved F1 scores of 0.31-0.57, primarily attributable to limited sample size and high linguistic variability. Conclusion The epilepsy-specific Chinese clinical annotation schema developed in this study demonstrates broad coverage, fine granularity, and high inter-annotator consistency. It effectively supports the training and evaluation of NER models and provides a reusable corpus foundation for the structured analysis of epilepsy medical records and the development of downstream intelligent diagnostic and therapeutic tools.

关键词

癫痫 / 电子健康档案 / 自然语言处理 / 命名实体识别 / 语料库

Key words

Epilepsy / Electronic health records / Natural language processing / Named entity recognition / Corpus

参考文献

[1] Thijs R D, Surges R, O'brien T J, et al. Epilepsy in adults[J]. Lancet (London, England), 2019, 393(10172): 689. DOI: 10.1016/S0140-6736(18)32596-0.

[2] Ding D, Zhou D, Sander J W et al. Epilepsy in China: major progress in the past two decades[J]. Lancet Neurol, 2021, 20(4): 316. DOI: 10.1016/S1474-4422(21)00023-5.

[3] Boleti A P A, Cardoso P H O, Frihling B E F, et al. Pathophysiology to risk factor and therapeutics to treatment strategies on epilepsy [J/OL]. Brain Sciences, 2024, 14(1): 71. https://www.mdpi.com/2076-3425/14/1/71. DOI: 10.3390/brainsci14010071.

[4] Boßelmann C M. Seizures, semiology, and syndromes: a narrative review[J]. Seizure, 2021, 92: 230. DOI: 10.1016/j.seizure.2021.09.019.

[5] Riney K, Bogacz A, Somerville E, et al. International League Against Epilepsy classification and definition of epilepsy syndromes with onset at a variable age: position statement by the ILAE Task Force on Nosology and Definitions[J]. Epilepsia, 2022, 63(6): 1443. DOI: 10.1111/epi.17240.

[6] 中国抗癫痫协会. 临床诊疗指南: 2023修订版. 癫痫病分册[M]. 北京: 人民卫生出版社, 2023.

[7] Chinese Association Against Epilepsy. Clinical practice guidelines: 2023 revised edition. Epilepsy volume[M]. Beijing: People's Medical Publishing House, 2023.

[8] Bodenreider O. The Unified Medical Language System (UMLS): integrating biomedical terminology [J/OL]. Nucleic acids research, 2004, 32(Database issue): D267. https://academic.oup.com/nar/article/32/suppl_1/D267/2505235. DOI: 10.1093/nar/gkh061.

[9] Gaudet-Blavignac C, Foufi V, Bjelogrlic M, et al. Use of the Systematized Nomenclature of Medicine Clinical Terms (SNOMED CT) for processing free text in health care: systematic scoping review [J/OL]. J Med Int Res, 2021, 23(1): e24594. https://www.jmir.org/2021/1/e24594. DOI: 10.2196/24594.

[10] 杜晋华, 尹浩, 冯嵩. 中文电子病历命名实体识别的研究与进展[J]. 电子学报, 2022, 50(12): 3030. DOI: 10.12263/DZXB.20220485.

[11] Du J, Yin H, Feng S. Research and development of named entity recognition in Chinese electronic medical record[J]. Acta Electronica Sinica, 2022, 50(12): 3030. DOI: 10.12263/DZXB.20220485.

[12] Tao S, Lhatoo S, Hampson J, et al. A bespoke electronic health record for epilepsy care (EpiToMe): development and qualitative evaluation [J/OL]. J Med Int Res, 2021, 23(2): e22939. https://www.jmir.org/2021/2/e22939. DOI: 10.2196/22939.

[13] Tjong Kim Sang E F, De Meulder F. Introduction to the CoNLL-2003 shared task: language-independent named entity recognition [C]// Proceedings of CoNLL-2003. Edmonton: Association for Computational Linguistics, 2003: 142-147. DOI: 10.3115/1119176.1119195.

[14] Artstein R, Poesio M. Inter-coder agreement for computational linguistics[J]. Computational Linguistics, 2008, 34(4): 555. DOI: 10.1162/coli.07-034-R2.

[15] Jehangir B, Radhakrishnan S, Agarwal R. A survey on named entity recognition-datasets, tools, and methodologies [J/OL]. Natural Language Processing Journal, 2023, 3: 100017. https://www.sciencedirect.com/science/article/pii/S2949719123000146. DOI: 10.1016/j.nlp.2023.100017.

[16] Su J, Murtadha A, Pan S, et al. Global pointer: novel efficient span-based approach for named entity recognition [EB/OL]. arXiv: 2208.03054, 2022. https://arxiv.org/abs/2208.03054. DOI: 10.48550/arXiv.2208.03054.

[17] Geirhos R, Jacobsen J H, Michaelis C, et al. Shortcut learning in deep neural networks[J]. Nat Mach Intell, 2020, 2(11): 665. DOI: 10.1038/s42256-020-00257-z.

[18] Cui L, Bozorgi A, Lhatoo S D, et al. EpiDEA: extracting structured epilepsy and seizure information from patient discharge summaries for cohort identification [C]// AMIA Annual Symposium Proceedings. Chicago: American Medical Informatics Association, 2012: 1191-1200.

[19] Decker B M, Turco A, Xu J, et al. Development of a natural language processing algorithm to extract seizure types and frequencies from the electronic health record[J]. Seizure, 2022, 101: 48. DOI: 10.1016/j.seizure.2022.07.010.

[20] Reeves R M, Christensen L, Brown J R, et al. Adaptation of an NLP system to a new healthcare environment to identify social determinants of health[J/OL]. J Biomed Inform, 2021, 120: 103851. https://www.sciencedirect.com/science/article/pii/S1532046421001805. DOI: 10.1016/j.jbi.2021.103851.

[21] Xia Y, He M, Basang S, et al. Semiology extraction and machine learning-based classification of electronic health records for patients with epilepsy: retrospective analysis [J/OL]. JMIR Med Inform, 2024, 12: e57727. https://medinform.jmir.org/2024/1/e57727. DOI: 10.2196/57727.

[22] Biswal S, Nip Z, Moura Junior V, et al. Automated information extraction from free-text EEG reports [C]// 2015 37th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC). Milan: IEEE, 2015: 6804-6807. DOI: 10.1109/EMBC.2015.7319956.

[23] Xie K, Gallagher R S, Conrad E C, et al. Extracting seizure frequency from epilepsy clinic notes: a machine reading approach to natural language processing[J]. J Am Med Inform Assoc, 2022, 29(5): 873. DOI: 10.1093/jamia/ocac018.

[24] Abeysinghe R, Tao S, Lhatoo S D, et al. Leveraging pretrained language models for seizure frequency extraction from epilepsy evaluation reports [J/OL]. NPJ Digital Medicine, 2025, 8(1): 208. https://www.nature.com/articles/s41746-025-01592-4. DOI: 10.1038/s41746-025-01592-4.

[25] Deußer T, Hillebrand L, Bauckhage C, et al. Informed named entity recognition decoding for generative language models [C]// 2024 IEEE International Conference on Big Data (BigData). Washington, DC: IEEE, 2024: 6001-6010. DOI: 10.1109/BigData62323.2024.10825603.

[26] Yew A N J, Schraagen M, Otte W M, et al. Transforming epilepsy research: a systematic review on natural language processing applications[J]. Epilepsia, 2022, 64(2): 292. DOI: 10.1111/epi.17474.

[27] Man J, Hu Z, Liu H, et al. A model for epilepsy named entity recognition based on Chinese EEG reports [C]// Proceedings of the 2024 5th International Symposium on Artificial Intelligence for Medicine Science. Chengdu: Association for Computing Machinery, 2024: 467-472. DOI: 10.1145/3706890.3706971.

[28] Sung M, Jeong M, Choi Y, et al. BERN2: an advanced neural biomedical named entity recognition and normalization tool[J]. Bioinformatics, 2022, 38(20): 4837. DOI: 10.1093/bioinformatics/btac598.

[29] Yang H, Müller J, Eberlein M, et al. Seizure forecasting with ultra long-term EEG signals[J]. Clin Neurophysiol, 2024, 167: 211. DOI: 10.1016/j.clinph.2024.09.017.

[30] Jones F J S, Smith J R, Ayub N, et al. Implementing standardized provider documentation in a tertiary epilepsy clinic [J/OL]. Neurology, 2020, 95(2): e213. https://www.neurology.org/doi/10.1212/WNL.0000000000009778.

[31] Linehan C, Benson A, Gunko A, et al. Exploring the prevalence and profile of epilepsy across Europe using a standard retrospective chart review: Challenges and opportunities[J]. Epilepsia, 2021, 62(11): 2651. DOI: 10.1111/epi.17057.

[32] Khemani D B, Malave D S, Shinde S, et al. AI-driven pharmacovigilance: enhancing adverse drug reaction detection with deep learning and NLP [J/OL]. MethodsX, 2025, 15: 103460. https://www.sciencedirect.com/science/article/pii/S221501612500305X. DOI: 10.1016/j.mex.2025.103460.

[33] Powell J R, Cook J, Wang Y, et al. Drug dosing recommendations for all patients: a roadmap for change[J]. Clin Pharmacol Therap, 2021, 109(1): 65. DOI: 10.1002/cpt.1923.

[34] Beniczky S, Ryvlin P. Mobile health and digital technology in epilepsy: the dawn of a new era [J]. Epilepsia, 2023, 64(S4): S1. DOI: 10.1111/epi.17813.

[35] Nuthalapati P, Thomas L, Donahue M A, et al. Improving seizure frequency documentation and classification [J/OL]. Neurology, 2023, 13(6): e200212. https://www.neurology.org/doi/10.1212/CPJ.0000000000200212.

基金资助

*四川省重大科技专项项目(2024ZDZX0018)