| 引用本文: |
王朝雨,黄奎麟,代国威,强茗,王倩.基于卷积神经网络的脑卒中中医辨证分型舌象分类研究[J].湖南中医药大学学报,2023,43(8):1460-1467[点击复制] |
|
| |
|
|
| 本文已被：浏览 898次 下载 1098次 |
| 基于卷积神经网络的脑卒中中医辨证分型舌象分类研究 |
| 王朝雨,黄奎麟,代国威,强茗,王倩 |
| (广汉市中医医院, 四川 广汉 618399;中国农业科学院农业信息研究所国家农业科学数据中心, 北京 100081;四川省中医药科学院中医研究所针灸康复科, 四川 成都 610031;清华大学第一附属医院神经内科, 北京 100016) |
| 摘要: |
| 目的 通过卷积神经网络学习脑卒中中医辨证分型与中医舌象特征分类的关系，为探索新的脑卒中临床标准化治疗方法提供诊断依据。方法 本研究选取284名脑卒中患者作为研究对象，通过迁移学习微调改进DenseNet201用于特征向量的提取，使用信息增益、卡方检验、对称不确定性与ReliefF滤波算法并组合去重以选择特征向量，最后利用Cubic SVM形成交叉数据集在多种分类器上进行训练和测试，比较模型的准确性。结果 试验结果表明，组合的四类特征提取算法使得准确率高于基础结果的3.26%，Cubic SVM分类器相对于其他分类器以及未改进的DenseNet201取得了最优结果，可以在脑卒中中医舌象辨证分型中提供至少为95.74%的准确率。结论 本研究提出的TCM舌象分类模型的方法结构是有效的，可辅助临床中医师进行诊断治疗，值得临床推广和进一步深入研究。 |
| 关键词: 中医舌诊 深度学习 舌象分类 支持向量机 特征选择 |
| DOI：10.3969/j.issn.1674-070X.2023.08.019 |
| 投稿时间：2023-03-10 |
| 基金项目:国家重点研发计划项目（2021YFF0704200）；四川省中医药管理局中医药科研专项课题面上项目（2023MS385）；广汉市社会发展医疗卫生领域重点研发项目（GH2022SFY15）。 |
|
| Convolutional neural networks-based TCM pattern differentiation and classification of tongue manifestations in stroke patients |
| WANG Zhaoyu,HUANG Kuilin,DAI Guowei,QIANG Ming,WANG Qian |
| (Guanghan Hospital of Chinese Medicine, Guanghan, Sichuan 618399, China;Agricultural Information Institute, Chinese Academy of Agricultural Sciences, National Agriculture Science Data Center, Beijing 100081, China;Department of Acupuncture and Rehabilitation, Institute of Chinese Medicine, Sichuan Academy of Chinese Medicine, Chengdu, Sichuan 610031, China;Department of Neurology, the First Hospital of Tisnghua University, Beijing 100016, China) |
| Abstract: |
| Objective To learn the relationship between TCM pattern differentiation and feature classification of TCM tongue manifestations in stroke patients through convolutional neural networks (CNN), so as to provide a diagnostic basis for exploring new standardized clinical treatment methods for stroke. Methods A total of 284 stroke patients were selected as subjects, and DenseNet201 was fine-tuned by transfer learning for feature vector extraction. The feature vectors were selected using information gain, chi-square test, symmetric uncertainty, and ReliefF filtering algorithm, combined with deduplication. Finally, Cubic SVM was used to form a crossover dataset for training and testing on multiple classifiers to compare the accuracy of the models. Results The combined four-class feature extraction algorithm resulted in an accuracy rate higher than 3.26% of the basic result, and compared with other classifiers and the unimproved DenseNet201, the Cubic SVM classifier achieved optimal results, providing an accuracy rate of at least 95.74% in the pattern differentiation of TCM tongue manifestations in stroke patients. Conclusion The methodological structure for the classification model of TCM tongue manifestations proposed in this study is effective, and it can assist clinical TCM physicians in diagnosis and treatment, which is worthy of clinical promotion and further in-depth research. |
| Key words: TCM tongue diagnosis deep learning classification of tongue manifestations support vector machine feature selection |
|
二维码（扫一下试试看！） |
|
|
|
|
