基于决策树算法的卒中后抑郁肝郁脾虚证诊断模型研究

The study on the diagnosis model for post-stroke depression with the pattern of liver depression and spleen deficiency based on decision tree algorithm

关键词:  卒中后抑郁  肝郁脾虚证  决策树  机器学习  诊断模型  证候诊断

Key words:  post-stroke depression  the pattern of liver depression and spleen deficiency  decision tree  machine learning  diagnostic model  pattern diagnosis

摘要:

[目的] 基于决策树算法建立卒中后抑郁肝郁脾虚证诊断模型，并评价与验证模型的诊断性能。[方法] 从临床病例中收集患者基本信息及中医证候类型数据，运用CART、CHAID、QUEST及C5.0算法建立卒中后抑郁肝郁脾虚证诊断模型，并使用准确率、AUC等指标在内部测试集及外部验证数据集上对模型性能进行综合评估。[结果] 建立CART、CHAID、QUEST及C5.0决策树模型，其中C5.0决策树模型的性能最优，准确率为90.91%，AUC为0.93。得到“胁肋胀痛”“神疲懒言”“腹胀”症状为诊断卒中后抑郁肝郁脾虚证最核心的分类特征，“神疲懒言”症状对判断卒中后抑郁患者是否属肝郁脾虚证具有最强的区分能力。[结论] 通过决策树算法建立卒中后抑郁肝郁脾虚证诊断模型具有较高的准确率，可以为卒中后抑郁肝郁脾虚证的诊断及临床表现规律的探索提供参考。

Abstract:

[Objective] To establish a diagnostic model for post-stroke depression with the pattern of liver depression and spleen deficiency using decision tree algorithms，and to evaluate its diagnostic performance. [Methods] Clinical case data were collected to obtain basic patient information and data on traditional Chinese medicine syndrome patterns. Diagnostic models for post-stroke depression with the pattern of liver depression and spleen deficiency were constructed using the CART，CHAID，QUEST，and C5.0 decision tree algorithms. The performance of these models was then evaluated on an internal test set and an external validation dataset using metrics including accuracy and Area Under the Curve（AUC）. [Results] Diagnostic models were successfully established using the CART，CHAID，QUEST，and C5.0 decision tree algorithms. Among these，the C5.0 decision tree model demonstrated the best performance，achieving an accuracy of 90.91 percent and an AUC of 0.93. The symptoms of “distending pain in the hypochondria” “lack of energy and reluctance to speak” and “abdominal distension” were identified as the core classification features for diagnosing post-stroke depression with the pattern of liver depression and spleen deficiency. Among these，the symptom “lack of energy and reluctance to speak” exhibited the strongest discriminative ability for determining whether patients with post-stroke depression exhibited the pattern of liver depression and spleen deficiency. [Conclusion] The diagnostic model for post-stroke depression with the pattern of liver depression and spleen deficiency based on decision tree algorithms demonstrates high accuracy and can provide a reference for the diagnosis of post-stroke depression with the pattern of liver depression and spleen deficiency and the exploration of its clinical presentation patterns.