Abstract:

Non-alcoholic fatty liver disease (NAFLD) is the most common chronic liver disease, with an increasing incidence in recent years. The current clinical strategy for managing the health of NAFLD risk groups adopts a “one-size-fits-all” approach, disregarding the difference among various risk stratification patients. Therefore, this paper proposed a framework for stratified health management and multi-dimensional metabolic index data mining method for NAFLD-risk people. First, it measured the metabolic index time series data of the patients using the dynamic temporal warping method, and utilized the similarity network fusion algorithm to combine the similarity matrix of multiple metabolic-related indicators. Next, it constructed a patients’ metabolism-related indicators social network, and used the Leiden community detection algorithm for risk stratification. Then, it conducted knowledge discovery using the structured and unstructured textual data of each sub-community, mining typical patterns, TOP-N keywords, and medical recommendations. Finally, it applied the framework to real-world datasets and developed prototype systems. It is found that the framework and method proposed in this paper can enable clinicians to better grasp the risk level and accompanying characteristics of the risk people, thus implementing more refined health management with a stratified medical approach.

Key words:

non-alcoholic fatty liver disease (NAFLD), multi-dimensional metabolic index, stratified health management, community detection

摘要：

非酒精性脂肪肝病（NAFLD）是最常见的慢性肝病，近年来其患病率不断增加。目前临床上针对 NAFLD 风险人群健康管理策略采用“一体适用”的方法，忽视了不同危险分层患者之间的差异性，为此，提出NAFLD风险人群分层健康管理框架及多维代谢数据挖掘方法。首先，使用动态时间扭曲算法对受检者的多维代谢指标时间序列数据进行相似性度量，并使用相似性网络融合算法对多维代谢指标时间序列相似度矩阵进行融合；其次，构建受检者多维代谢指标社交网络，通过Leiden社区发现算法进行危险分层；再次，基于各个子社区的结构化数值型数据与非结构化文本型数据进行知识发现，挖掘典型模式与TOP-N主题词及健康建议；最后，将该框架应用于真实世界数据集并开发原型系统。实验结果表明，本文提出的NAFLD风险人群分层健康管理框架及方法能够使临床医生更好地了解受检者危险层级及相应特征，从而以分层医学思想实施更精细化的健康管理。

关键词:

非酒精性脂肪肝, 多维代谢指标, 分层健康管理, 社区发现