結合醫學專業知識幫助醫療編碼預測

Incorporating Medical Domain Knowledge into Clinical Code Prediction

黃敏(Min Huang)

指導教授 ： 鄭卜壬

國立臺灣大學/電機資訊學院/資訊工程學研究所/碩士(2021年)

https://doi.org/10.6342/NTU202102125

摘要

醫療編碼指為醫學敘述加上編碼，用以表示醫療診斷和處置。它的好處在於將自由形式的文字標準化，可應用於：健康追蹤、醫療決策、統計分析、保險費用估價等。其中，最常用的國際編碼以國際疾病統計分類(ICD)為大宗。現今在醫院內，有疾病分類師為病歷標上國際疾病統計分類；然而，國際疾病統計分類的類別眾多，即使是專業人員也不一定能正確標記。隨著電子病歷的普及和預測模型的發展，自動化國際疾病統計分類成為一個長遠的研究目標。
在這篇論文中，我們提出了一個方法結合醫學知識圖譜以幫助國際疾病統計分類。我們的核心想法是：病歷中重要的醫學文字應對分類結果造成較大的影響。我們提取病歷中的醫學概念，透過計算概念和國際疾病統計分類的相似度，藉此提高重要醫學文字在注意力機制中的權重。在知識圖譜的幫助下，實驗顯示我們提出的方法能幫助先前的模型更好地預測國際疾病統計分類。

關鍵字

醫療編碼預測 ； 國際疾病分類 ； 注意力機制 ； 知識圖譜嵌入

並列摘要

Clinical coding refers to translate medical narratives to code representation for indicating medical diagnosis and procedures. Its advantage lies in standardizing free-text, which can be applied to health tracking, medical decision-making, statistical analysis, insurance pricing, etc. Among them, International Statistical Classification of Diseases and Related Health Problems (ICD) is the most commonly used. Nowadays, clinical coders label ICD in hospitals. However, there are so many categories in the ICD ontology that even professionals might not be able to label them correctly. With the development of electronic medical records and prediction models, automatic ICD coding has become a long-term research goal.

In this paper, we propose a method to combine medical knowledge graph to improve ICD coding models. Our main idea is that important medical words in clinical texts should contribute more to the prediction results. We extract the medical concepts from the clinical texts. Then, we calculate the concept similarity between these concepts and ICD to increase the weights of important medical words in the attention mechanism. Experiments show that our approach helps the previous model better predict the ICD.

並列關鍵字

Medical Code Prediction ； ICD Coding ； Attention Mechanism ； Knowledge Graph Embedding

參考文獻

[1] Xuedong Li, Yue Wang, Dongwu Wang, Walter Yuan, Dezhong Peng, and Qiaozhu Mei. Improving rare disease classification using imperfect knowledge graph. BMC medical informatics and decision making, 19(5):1–10, 2019.

[2] Mohammed Alawad, Shang Gao, Mayanka Chandra Shekar, SM Hasan, J Blair Christian, Xiao­Cheng Wu, Eric B Durbin, Jennifer Doherty, Antoinette Stroup, Linda Coyle, et al. Integration of domain knowledge using medical knowledge graph deep learning for cancer phenotyping. arXiv preprint arXiv:2101.01337, 2021.

[3] Leah S Larkey and W Bruce Croft. Combining classifiers in text categorization. In Proceedings of the 19th annual international ACM SIGIR conference on Research and development in information retrieval, pages 289–297, 1996.

[4] Alistair EW Johnson, Tom J Pollard, Lu Shen, H Lehman Li­Wei, Mengling Feng, Mohammad Ghassemi, Benjamin Moody, Peter Szolovits, Leo Anthony Celi, and Roger G Mark. Mimic­iii, a freely accessible critical care database. Scientific data, 3(1):1–9, 2016.

[5] James Mullenbach, Sarah Wiegreffe, Jon Duke, Jimeng Sun, and Jacob Eisenstein. Explainable prediction of medical codes from clinical text. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers), pages 1101–1111, 2018.