基于EfficientNet的胃癌舌像变化分析研究  

Analysis of Tongue Image Changes in Gastric Cancer Based on EfficientNet

作　　者：周子力 胡文 邓巍 何绍亚 ZHOU Zili;HU Wen;DENG Wei(Department of Gastroenterology,The Second Hospital of Traditional Chinese Medicine of Sichuan Province,Chengdu Sichuan 610031,China)

机构地区：[1]四川省第二中医医院消化内科,四川成都610031

出　　处：《四川中医》2024年第9期78-82,共5页Journal of Sichuan of Traditional Chinese Medicine

基　　金：四川省科技计划重点研发项目(编号:2023YFS0327,课题名称:消化系统中医优势病种中医智能舌诊设备研发)。

摘　　要：目的:通过结合EfficientNet模型研究胃癌舌像变化,分析舌色和苔色的类别,旨在使用便携式中医智能舌诊仪辅助医生做出更准确的诊断和治疗决策,经统计学技术分析不同舌色、苔色的量化指标。方法:采集5000例受试者舌像数据,对照组采用BP神经网络和卷积神经网络分别分类舌色和苔色。观察组结合EfficientNet模型辅助BP神经网络和卷积神经网络分别分类舌色和苔色,从而分析胃癌患者舌像的变化特征,并采用评价指标评价模型效果。结果:观察组模型研究胃癌舌像变化,精确率为96.59%。测试集中识别苔色召回率、精确率、准确率分别为92.63%、92.62%、90.60%。识别舌色召回率、精确率、准确率分别为89.61%、88.62%、88.45%。结论:EfficientNet网络模型可以提高识别舌色和苔色变化的效率,这为提高临床医生诊疗决策效率提供了有力支持。Objective By combining the EfficientNet model to study the changes in tongue images of gastric cancer,analyzing the categories of tongue color and coating color,the aim is to use a portable traditional Chinese medicine intelligent tongue diagnostic instrument to assist doctors in making more accurate diagnosis and treatment decisions.Statistical techniques are used to analyze quantitative indicators of different tongue and coating colors.Method Collect tongue image data from 5000 subjects,and use BP neural network and convolutional neural network to classify tongue color and coating color in the control group,respectively.The observation group combined EfficientNet model with BP neural network and convolutional neural network to classify tongue color and coating color respectively,in order to analyze the changes in tongue images of gastric cancer patients,and used evaluation indicators to evaluate the effectiveness of the model.Result The observation group model was used to study the changes in tongue images of gastric cancer,with an accuracy rate of 96.59%.The recall rate,accuracy rate,and accuracy rate of identifying moss color in the test set are 92.63%,92.62%,and 90.60%,respectively.The recall rate,accuracy rate,and accuracy rate of tongue color recognition are 89.61%,88.62%,and 88.45%,respectively.Conclusion The EfficientNet network model can improve the efficiency of identifying changes in tongue color and coating color,which provides strong support for improving the efficiency of clinical doctors in diagnosis and treatment decision-making.

关 键 词：胃癌 舌像变化 EfficientNet 舌色 苔色 精确率 

分 类 号：R241.25[医药卫生—中医诊断学]