基于深度学习ConvNeXt模型的冠心病痰湿证舌诊信息分类辨识

Classification and identification of tongue diagnosis information of phlegm-dampness syndrome of coronary heart disease based on deep learning ConvNeXt model

目的通过卷积神经网络对冠心病痰湿证的舌象进行分类识别,提高冠心病痰湿证舌象识别准确率。方法选取2020年10月至2022年8月在内蒙古自治区乌兰浩特市妇幼保健院、陕西中医药大学第二附属医院等地采集到的200例冠心病患者,其中痰湿证组、非痰湿证组各100例,运用ConvNeXt模型、朴素贝叶斯网络、K近邻模型、决策树算法、支持向量机模型进行舌象分类辨识。结果不同模型的舌象分类平均准确度均在50%以上,ConvNeXt模型的平均准确度最高为89.44%;ConvNeXt模型验证集中痰湿证和非痰湿证2个类别的平均准确度、精确率、F1值和召回率均接近90%。结论使用ConvNeXt模型进行舌象分类识别,能够较为准确地从舌诊上区分冠心病痰湿证与非痰湿证,客观化的人工智能识别技术,可以辅助冠心病痰湿证的临床诊断,有助于中医舌诊客观化研究的发展。

Objective To classify and identify the tongue image of phlegm-dampness syndrome of coronary heart disease by convolutional neural network,so as to improve the recognition accuracy of tongue image of coronary heart disease.Methods From October 2020 to August 2022,200 patients with coronary heart disease were collected from Ulanhot Maternal and Child Health Hospital in Inner Mongolia Autonomous Region,and the Second Affiliated Hospital of Shaanxi University of Chinese Medicine and so on,including 100 patients in the phlegm-dampness syndrome group and 100 patients in the non-phlegm-dampness syndrome group.ConvNeXt model,naive Bayesian network,K nearest neighbor,decision tree algorithm,and support vector machine model were used to classify and identify tongue image.Results The average accuracy of tongue image classification of different models was above 50%,and the average accuracy of ConvNeXt model was 89.44%.ConvNeXt model verified that the average accuracy,accuracy,F1 value,and recall rate of the two categories of concentrated phlegm-dampness syndrome and non-phlegm-dampness syndrome were close to 90%.Conclusion Tongue image classification and recognition using ConvNeXt model can more accurately distinguish phlegm-dampness syndrome of coronary heart disease from non-phlegm-dampness syndrome in tongue diagnosis.Objectified artificial intelligence recognition technology can assist clinical diagnosis of phlegm-dampness syndrome of coronary heart disease and contribute to the development of objectified research on tongue diagnosis of traditional Chinese medicine.