临床医学文本NLP数据集

328,000条电子病历、诊断报告、医嘱、文献摘要的NER与关系抽取标注数据集,支持临床NLP与术语标准化模型训练。

数据集概览

数据集名称临床医学文本 NLP 标注数据集
数据总量328,000 条(电子病历 + 诊断报告 + 医嘱 + 医学文献摘要)
文本类型入院记录 85K / 病程记录 72K / 出院小结 45K / 检查报告 63K / 医嘱 38K / 文献摘要 25K
NER 实体类别疾病、症状、药品、检查、手术、解剖部位、实验室指标 7大类 46子类
关系标注疾病-症状、药品-疾病、检查-发现、手术-解剖 4种关系类型
术语标准化ICD-10-CM / SNOMED CT / ATC / LOINC 映射
标注方式临床医学专业人员标注 + NLP预标注辅助 + 术语专家审核
格式输出BIO/BIOS 序列标注 / JSON / Brat Standoff / CoNLL

数据维度详细说明

字段名称数据类型取值范围/说明
doc_idString(32)脱敏文档ID
doc_typeEnumADMISSION | PROGRESS | DISCHARGE | REPORT | ORDER | LITERATURE
departmentEnumINTERNAL | SURGERY | OBGYN | PEDIATRICS | ONCOLOGY | CARDIOLOGY 等
textText脱敏后的临床文本(患者姓名→[NAME],日期→[DATE],机构→[ORG])
entitiesArray[{start, end, type, subtype, text, icd10_code, snomed_code}]
relationsArray[{head_entity_idx, tail_entity_idx, relation_type}]
bio_tagsArrayBIO/BIOS 序列标注标签列表
text_lengthInteger字符数(含标点)
entity_countInteger实体数量
annotator_idString(8)脱敏标注者ID
qc_scoreFloat质检评分 0.0 - 1.0

脱敏 JSON 数据样例

[ { "doc_id": "CLT-A0000821", "doc_type": "ADMISSION", "department": "CARDIOLOGY", "text": "患者[NAME],[AGE]岁[GENDER],因"反复胸闷、心悸[DURATION]"入院。既往有[DISEASE]病史[DRUG]治疗中。入院查体:[VALUE]次/分,血压[VALUE]/[VALUE]mmHg。", "entities": [ {"start": 62, "end": 70, "type": "SYMPTOM", "subtype": "胸闷", "text": "反复胸闷、心悸"}, {"start": 82, "end": 90, "type": "DISEASE", "subtype": "心血管疾病", "text": "[DISEASE]", "icd10_code": "I25.1"}, {"start": 92, "end": 98, "type": "DRUG", "subtype": "心血管药物", "text": "[DRUG]", "atc_code": "C07AB"} ], "relations": [ {"head_entity_idx": 0, "tail_entity_idx": 1, "relation_type": "SYMPTOM_OF"} ], "text_length": 156, "entity_count": 3, "annotator_id": "ANN-0421", "qc_score": 0.96 }, { "doc_id": "CLT-R0012345", "doc_type": "REPORT", "department": "RADIOLOGY", "text": "检查部位:[BODY_PART]。影像所见:[BODY_PART]见[SIZE]cm[DESCRIPTOR][FINDING],边界[DEGREE],周边[DESCRIPTOR]。诊断意见:[DIAGNOSIS]。", "entities": [ {"start": 4, "end": 16, "type": "ANATOMY", "subtype": "检查部位", "text": "[BODY_PART]", "snomed_code": "123037004"}, {"start": 20, "end": 32, "type": "ANATOMY", "subtype": "检查部位", "text": "[BODY_PART]"}, {"start": 33, "end": 42, "type": "LAB", "subtype": "尺寸测量", "text": "[SIZE]cm"}, {"start": 43, "end": 55, "type": "FINDING", "subtype": "影像发现", "text": "[DESCRIPTOR][FINDING]"} ], "relations": [ {"head_entity_idx": 1, "tail_entity_idx": 3, "relation_type": "HAS_FINDING"} ], "text_length": 89, "entity_count": 4, "annotator_id": "ANN-0715", "qc_score": 0.94 }, { "doc_id": "CLT-D0003672", "doc_type": "DISCHARGE", "department": "ONCOLOGY", "text": "出院诊断:[DISEASE]([STAGE])。住院期间行[PROCEDURE]治疗,过程顺利。出院医嘱:1.[DRUG] [DOSAGE];2.定期复查[LAB];3.不适随诊。", "entities": [ {"start": 5, "end": 15, "type": "DISEASE", "subtype": "肿瘤", "text": "[DISEASE]", "icd10_code": "C34.9"}, {"start": 24, "end": 32, "type": "PROCEDURE", "subtype": "治疗操作", "text": "[PROCEDURE]"}, {"start": 47, "end": 53, "type": "DRUG", "subtype": "抗肿瘤药", "text": "[DRUG]", "atc_code": "L01XE"}, {"start": 64, "end": 70, "type": "LAB", "subtype": "实验室检查", "text": "[LAB]", "loinc_code": "24331-1"} ], "relations": [ {"head_entity_idx": 0, "tail_entity_idx": 1, "relation_type": "TREATED_BY"}, {"head_entity_idx": 0, "tail_entity_idx": 3, "relation_type": "MONITORED_BY"} ], "text_length": 112, "entity_count": 4, "annotator_id": "ANN-0388", "qc_score": 0.97 }, { "doc_id": "CLT-O0009105", "doc_type": "ORDER", "department": "INTERNAL", "text": "[DRUG] [DOSAGE] [ROUTE] [FREQUENCY] [DURATION]天", "entities": [ {"start": 0, "end": 6, "type": "DRUG", "subtype": "抗生素", "text": "[DRUG]", "atc_code": "J01MA"}, {"start": 7, "end": 15, "type": "DOSAGE", "subtype": "剂量", "text": "[DOSAGE]"}, {"start": 16, "end": 22, "type": "ROUTE", "subtype": "给药途径", "text": "[ROUTE]"} ], "relations": [], "text_length": 42, "entity_count": 3, "annotator_id": "ANN-0912", "qc_score": 0.99 }, { "doc_id": "CLT-L0001543", "doc_type": "LITERATURE", "department": "ONCOLOGY", "text": "本研究旨在评估[DISEASE]患者中[DRUG]联合[DRUG]方案的有效性和安全性。共纳入[NUMBER]例患者,ORR为[VALUE]%,中位PFS为[VALUE]个月。", "entities": [ {"start": 6, "end": 16, "type": "DISEASE", "subtype": "肿瘤", "text": "[DISEASE]"}, {"start": 18, "end": 24, "type": "DRUG", "subtype": "抗肿瘤药", "text": "[DRUG]"}, {"start": 26, "end": 32, "type": "DRUG", "subtype": "抗肿瘤药", "text": "[DRUG]"}, {"start": 40, "end": 48, "type": "LAB", "subtype": "临床指标", "text": "ORR"}, {"start": 55, "end": 63, "type": "LAB", "subtype": "临床指标", "text": "中位PFS"} ], "relations": [ {"head_entity_idx": 0, "tail_entity_idx": 1, "relation_type": "TREATED_BY"}, {"head_entity_idx": 0, "tail_entity_idx": 3, "relation_type": "MEASURED_BY"} ], "text_length": 97, "entity_count": 5, "annotator_id": "ANN-0523", "qc_score": 0.95 } ]

标注规范与质量控制

标注标准

  • NER 标注遵循 i2b2 2010/2012 临床命名实体识别标准扩展,新增 46 个细粒度子类别
  • 关系抽取采用「头实体→关系→尾实体」三元组格式,关系类型覆盖诊断-发现、治疗-疾病等 4 大类
  • 术语标准化双映射:ICD-10-CM 诊断编码 + SNOMED CT 临床术语,药品用 ATC 编码
  • 文本脱敏:PHI(Protected Health Information)全部替换为类型占位符,零信息泄露

质量指标

96.3%

NER F1

91.7%

关系抽取F1

98.5%

术语映射准确率

0.89

标注一致率Kappa

AI 训练适用场景

临床命名实体识别(NER)

适用模型:BERT-BiLSTM-CRF / BioBERT / PubMedBERT / ClinicalBERT / GatorTron。训练面向中文临床文本的命名实体识别模型,F1 可达 90%+。

医学关系抽取

适用模型:BioRelEx / PURE / PL-Marker / REBEL。在实体识别基础上抽取疾病-症状、药品-疾病等关系,构建医学知识图谱。

临床术语标准化

适用模型:SapBERT / CODER / BioSyn。将口语化/非标准化临床术语映射到 ICD-10/SNOMED CT 标准术语体系。

电子病历结构化

适用模型:LLM + RAG(GPT-4o/Claude/Qwen2.5 + LangChain)。端到端将非结构化病历转为结构化 FHIR/CDA 格式。

认证专家网络

5000+
认证科研专家

覆盖数学、物理、化学、生物、地学、计算机六大学科

四重
质量审核体系

学术资质审核、专业能力评估、数据标注培训、质量持续监控

AI+
人机协同标注

AI预标注+专家复核,效率提升300%,准确率99.5%+

获取本数据集

临床NLP数据集按科室/标注类型灵活授权,支持定制化标注需求。

数据样本预览 · 定制化数据方案 · 专业技术支持

商业应用价值

智慧医院信息化

嵌入HIS/EMR系统中的智能病历质控、CDSS临床决策支持模块,提升病历规范率和诊疗质量。

医保智能审核

基于诊断-药品-检查关系抽取,训练医保智能审核模型,自动识别不合理用药和过度检查。

药企真实世界研究

从海量电子病历中结构化提取用药信息和临床结局,支撑药物上市后真实世界证据(RWE)研究。