基于卷积神经网络建立中药材自动识别的人工智能模型及应用程序  

Construction of an Artificial Intelligence Model and Application for an Automatic Recognition of Traditional Chinese Medicine Herbals Based on Convolutional Neural Networks

作　　者：王甘红 张子豪 奚美娟 夏开建 周燕婷 陈健[4] WANG Ganhong;ZHANG Zihao;XI Meijuan;XIA Kaijian;ZHOU Yanting;CHEN Jian(Department of Gastroenterology,Changshu Hospital of Traditional Chinese Medicine(Changshu New District Hospital),Changshu 215500,China;Shanghai Hao Brothers Educational Technology Co.,Ltd.,Shanghai 200434,China;Changshu Key Laboratory of Medical Artificial Intelligence and Big Data,Changshu 215500,China;Department of Gastroenterology,Changshu No.1 People's Hospital,Changshu 215500,China)

机构地区：[1]江苏省常熟市中医院(常熟市新区医院)消化内科,215500 [2]上海豪兄教育科技有限公司,上海市200434 [3]江苏省常熟市医学人工智能与大数据重点实验室,215500 [4]江苏省常熟市第一人民医院消化内科,215500

出　　处：《中国全科医学》2025年第9期1128-1136,共9页Chinese General Practice

基　　金：常熟市医学人工智能与大数据重点实验室能力提升项目(CYZ202301);常熟市医药卫生科技计划项目(CSWS202316);常熟市科技发展计划项目(CS202019);江苏省333高层次人才培养工程(SZFCXK202147);苏州市应用基础研究科技创新项目(SYWD2024059)。

摘　　要：背景传统中药材检测手段依赖主观经验,难以满足中药材在准确分类与鉴别方面的需求。目的基于卷积神经网络(CNN)开发一款能够自动识别163种中药材的人工智能模型及电脑端应用程序。方法2020年1月—2024年6月,采集了两个中药材数据集进行深度学习模型的训练、验证和测试,共包含163种中药材。通过准确率、灵敏度、特异度、精确率、受试者工作特征(ROC)曲线下面积(AUC)、F1分数等指标来衡量CNN模型的性能。在模型训练完成后,基于PyQt5技术开发了一款应用程序,供临床便携使用。结果本研究共纳入了276767张图像,开发了EfficientNetB0、ResNet50、MobileNetV3、VGG19和ResNet185种模型,通过性能比较,EfficientNet_B0模型在验证集上取得了最高的准确率(99.0%)和AUC(0.9942),被选为最佳模型。在测试集上,最佳模型对所有中药类别识别的准确率为99.0%、灵敏度为99.0%、特异度为100.0%、AUC为1.0,展现出良好的性能。结论基于卷积神经网络开发的深度学习模型能够快速准确地识别163种中药材,借助其高灵敏度的识别能力,为医师对中药材的鉴别提供有力辅助。Background Conventional methods for identifying traditional Chinese medicine(TCM)herbals mainly rely on subjective experiences,making it difficult to meet the needs for accurate classification and identification.Objective This study aims to develop an artificial intelligence model and a desktop application capable of automatically recognizing 163 types of TCM herbals based on convolutional neural networks(CNN).Methods From January 2020 to June 2024,data from two datasets of 163 TCM herbals were collected for training,validation,and testing of the deep learning model.The performance of the CNN model was evaluated for the accuracy,sensitivity,specificity,precision,area under the receiver operating characteristic(ROC)curve(AUC),and F1 score.After model training,an application was developed using PyQt5 technology for convenient clinical use.Results A total of 276767 images were included in this study.Five models,including EfficientNetB0,ResNet50,MobileNetV3,VGG19,and ResNet18,were developed.After comparing their performance,the EfficientNetB0 model achieved the highest accuracy(99.0%)and AUC(0.9942)in the validation dataset,and it was selected as the optimal model.In the test dataset,the EfficientNetB0 model achieved an accuracy of 99.0%,sensitivity of 99.0%,specificity of 100.0%,and an AUC of 1.0 across all categories,demonstrating an excellent performance.Conclusion The deep learning model developed based on CNN can quickly and accurately recognize 163 types of TCM herbals with high sensitivity and recognition capability,thus providing a robust support for physicians to accurately identify TCM herbals.

关 键 词：中药材 模式识别 自动 中药药材学 应用程序 人工智能 PyQt5 卷积神经网络 

分 类 号：R282[医药卫生—中药学]