基于卷积神经网络的新冠肺炎CT图像识别系统

COVID-19CT Image Recognition System Based on Convolutional Neural Network

新型冠状病毒肺炎具有以其传播速度快、传染率高的特点,使其成为全世界发病率和死亡率都极高的急性呼吸道传染病,肺部肺结节异常状态的正确识别和新冠肺炎是否感染的准确判别对于全世界医学发展与计算机算法逻辑都具有极其重要的发展意义。本项目所使用的卷积神经网络(CNN)就是通过计算机算法神经网络模拟大脑的学习分辨过程,先从识别图像数据基层初步提取特征,再用神经网络对基层特征进行概括与再聚合。系统以新冠肺炎CT图像为例,采用COVID-19 CHEST X-RAY DATABASE数据集,包含阳性1710张、阴性1345张。为提高检测精度采取了迁移学习的方法进行卷积神经网络的二次深度学习。通过实验得到,构建的模型可以完成是否患有新冠肺炎的分类判断,新冠患者CT影像的平均识别率可达92%,非新冠患者CT影像的平均识别率可达96%。

COVID-19’ characteristic is rapid spread and high infection rate,making it the world extremely high morbidity and mortality of acute respiratory infectious diseases,pulmonary lung nodule abnormal state correctly identify whether or not with COVID-19 infected accurate criterion for the world medicine development and the development of the computer algorithm logic has extremely important significance.The convolutional neural network(CNN) used in this project is to simulate the learning and resolution process of the brain through the computer algorithmic neural network.Features are initially extracted from the recognition image database,and then the neural network is used to summarize and reaggregate the basic features.In this system,COVID-19CT images are taken as an example and the COVID-19 CHEST X-ray DATABASE data set is used,including 1710 positive images and 1345 negative images.To improve the detection accuracy,the transfer learning method is adopted for secondary deep learning of convolutional neural networks.Through experiments,the constructed model can complete the classification and judgment of WHETHER COVID-19 patients have COVID-19.The average recognition rate of CT images of COVID-19 patients can reach 92%,and that of non-COVID-19 patients can reach 96%.