Abstract: Abstract: Concern of structural diversity of medical Electronic Health Records (EHRs) and the complicated correlation between coding in the automatic ICD coding task, an automatic ICD method AIC-HI (Automatic ICD Coding integrating Heterogeneous Information) that integrates heterogeneous information was proposed. Firstly, various feature extractors were designed based on the distinctive characteristics of structured coding, semi-structured description, and unstructured medical text in the coding task; At the same time, the coding knowledge graph was constructed to fit the hierarchical relationship of coding, and the association relationship between different branches was transformed into a triple containing head and tail coding; Then representation learning was used to fuse encoding and description information to calculate label features; Finally, the attention mechanism was used to extract the most relevant feature representation in unstructured documents. The experimental results show that, compared with the suboptimal baseline model MARN (Multitask bAlanced and Recalibrated Network), the microscopic F1-score of the model AIC-HI on the real clinical dataset MIMIC-III is increased by 4.3 percentage points.

Key words: Keywords: Medical Code Prediction, Automatic ICD Coding, Hierarchical Structure, Heterogeneous Information, Natural Language Processing (NLP)

摘要： 摘 要: 针对自动ICD编码中医学电子健康记录的结构多样性以及编码间复杂的关联关系等特点，提出了一种融合异构信息的自动ICD编码方法AIC-HI (Automatic ICD coding integrating heterogeneous information)。具体而言，首先针对编码任务中结构化编码、半结构化描述、非结构化医学文本三种异构数据的不同特性设计了多种特征提取器；同时构建编码知识图谱来拟合编码的层次结构关系，将不同分支间关联关系转化为包含头尾编码的三元组；然后运用表征学习融合编码和描述信息计算出标签特征；最后通过注意力机制提取在非结构化文档中与编码标签最为相关的特征表示。实验结果表明，与次优的基线模型MARN (Multitask bAlanced and Recalibrated Network) 相比，模型AIC-HI在真实临床数据集MIMIC-III上所有编码的微观F1值提升了4.3个百分点。

关键词: 关键词: 医学代码预测, 自动ICD编码, 层次结构, 异构信息, 自然语言处理