分层可解释机器学习模型用于穴位处方确定
本研究用机器学习方法基于症状为患者生成个体化穴位处方,以提升针灸疗效与效率。先构建穴位处方数据库,选用分层注意力循环神经网络(HARNN)建模,5折交叉验证并以交并比(IoU)为主要指标评价。原始5000例数据集上IoU为0.883,数据增强后达0.954,1000例原始样本测试集为0.932,并用LIME可视化预测依据。
要点
- HARNN可依据症状输出穴位处方,原始数据集IoU为0.883
- 数据增强后5折交叉验证IoU达0.954,测试集为0.932
- LIME可视化预测依据,提升模型临床可解释性
核心结论
- 研究构建了分层且可解释的机器学习模型,可按症状预测穴位处方,为针灸智能化提供技术路径与方法学参考。
研究设计与样本
- 研究先预处理针灸文本数据,构建面向机器学习应用的穴位处方数据库,并据此选择分层注意力循环神经网络(HARNN)建模。
- 采用5折交叉验证开展计算实验,以交并比(IoU)为主要评价指标,并应用局部可解释模型无关解释(LIME)方法可视化预测结果。
主要发现
- 在原始5000例样本数据集上,HARNN预测穴位处方的IoU为0.883。
- 数据增强后,5折交叉验证中IoU达到0.954。
- 在1000例原始样本测试集上,IoU为0.932。
- 使用LIME可直观可视化模型的预测依据,从而增强模型可靠性。
临床怎么用
- 该模型可用于根据患者症状生成个体化穴位处方,辅助提高针灸治疗效率与效果。
- LIME可视化预测依据,有助于临床医生理解并审阅模型给出的穴位选择。
- 目前为方法学研究,临床落地前仍需进一步验证。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
OBJECTIVE: This study used machine learning methods to develop a model that can offer personalized acupoint prescriptions for patients based on their symptoms, enhancing both the efficiency and effectiveness of acupuncture and moxibustion therapy (AMT). METHODS: We first preprocessed textual AMT data to build an acupoint prescription database designed for machine learning applications. Then, based on data analysis, we selected the hierarchical classification model hierarchical attention-based recurrent neural network (HARNN) to determine acupoint prescriptions based on symptoms. Computational experiments were conducted using 5-fold cross-validation to evaluate the model's performance, with intersection over union (IoU) as the primary evaluation metric. Finally, to enhance model interpretability, the local interpretable model-agnostic explanation (LIME) method was applied to visualize prediction results and improve its clinical applicability. RESULTS: On the original dataset of 5000 samples, HARNN achieved an IoU of 0.883 in predicting acupoint prescriptions. After data augmentation, the IoU reached 0.954 in 5-fold cross-validation, and 0.932 on a test set of 1000 original samples. The use of LIME enabled intuitive visualization of the model's prediction rationale, thereby enhancing the model's reliability. CONCLUSION: This study developed a hierarchical and interpretable machine learning model that predicts acupoint prescriptions based on symptoms, integrating HARNN for hierarchical classification and LIME for interpretability, which provides an effective technical approach and methodology for the intellectualization of AMT. Please cite this article as: Yang H, Wu R, Nakata M, Ge QW. A hierarchical and interpretable machine learning model for acupoint determination. J Integr Med. 2026; 24(3):340-351.
引用
- Yang H, Wu R, Nakata M, et al. A hierarchical and interpretable machine learning model for acupoint determination. J Integr Med. 2026. PMID: 41620365. doi:10.1016/j.joim.2026.01.003
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。