Intelligent Syndrome Differentiation of Traditional Chinese Medicine by ANN: A Case study of Chronic Obstructive Pulmonary Disease
Intelligent Syndrome Differentiation of Traditional Chinese Medicine by ANN: A Case study of Chronic Obstructive Pulmonary Disease
复制标题
人工神经网络中医智能辨证——以慢性阻塞性肺疾病为例
DOI：
10.1109/access.2019.2921318
复制
发表时间：
2019-01-01
期刊：
影响因子：
3.9
通讯作者：
Guo, Jinhong
中科院分区：
文献类型：
--
作者：
Xu, Qiang;Tang, Wenjun;Guo, Jinhong
中医药对于预防和治疗各种疾病具有良好的效果，已被纳入世界卫生组织（WHO）最新的全球医学纲要（2019版）。辨证论治作为中医最重要的特点之一，为疗效提供了保证。 SD是一个高维复杂函数，以症状/体征作为输入，以症状类型作为输出。人工神经网络 (ANN) 提供了一种通用的数据驱动解决方案来拟合高维复杂函数，使 ANN 成为中医智能 SD (ISD) 建模的有前景的方法。在本文中，我们选择慢性阻塞性肺疾病（COPD）作为基于人工神经网络（ANN）研究中医 ISD 的例子。首先，我们建立了一个全组 ANN 模型，将 ANN 与由 18471 条真实临床记录组成的全组数据集相结合。此外，我们还使用 ANN 和四个子组数据集构建了四个额外模型。为了进行比较，我们使用四种传统机器学习算法和全组数据集构建了另外四个模型。我们使用准确度和 F1 分数来评估模型的性能。全组ANN模型的准确率为86.45%，F1分数为82.93%，优于传统机器学习算法构建的四个比较模型，然而，四个子组模型比全组ANN模型取得了更好的性能。我们的结论是，人工神经网络有可能为中医 ISD 提供一种方法，并且我们的子组建模提出了进一步优化 ISD 的想法。
Traditional Chinese medicine (TCM) is effective in preventing and treating all manner of diseases, which has been incorporated into the latest global medical outline (Ver.2019) by World Health Organization (WHO). As one of the most important characteristics of TCM, syndrome differentiation (SD) provides curative effect assurance. SD is a high-dimensional complex function with symptoms/signs as input and syndrome type as output. Artificial neural network (ANN) provides an all-purpose data-driven solution to fit high-dimensional complex function, making ANN a promising approach for modeling intelligent SD (ISD) for TCM. In this paper, we chose chronic obstructive pulmonary disease (COPD) as an example for investigating ISD for TCM based on ANN. First, we built a full-group ANN model that combines ANN with full-group datasets composed of 18471 real clinical records. In addition, we built four extra models with ANN and four subgroup datasets. For comparison, we built another four models with four traditional machine-learning algorithms and the full-group datasets. We used accuracy and F1-scores to evaluate the models’ performance. With an accuracy of 86.45% and an F1 score of 82.93%, the full-group ANN model outperformed the four comparison models built from traditional machine-learning algorithms, and however, the four subgroup models achieved a better performance than the full-group ANN model. We concluded that the ANN can potentially provide a way for ISD for TCM, and our subgroup modeling suggests ideas for further optimizing the ISD.