基于卷积神经网络的肺炎检测系统

Pneumonia detection system based on convolutional neural network

提出一种利用卷积神经网络对胸部X光影像数据进行分析,实现对肺炎这一常见症状进行检测的系统。该方案首先进行图像预处理,然后采用VGG卷积神经网络和改进VGG卷积神经网络分别实现对胸部X光影像的特征提取,得到两种网络对肺炎的检测准确率,最终测试结果显示,改进VGG卷积神经网络在对肺炎检测上具备比VGG网络更好的识别效果,同时,改进VGG网络在标定好的数据集进行训练时能够更快的收敛。通过在公开数据集上进行测试表明,改进VGG卷积神经网络得到了98.5%的准确率,相比于VGG网络提升了2%以上的识别准确率,证明该方案在肺炎识别上具有可行性。

A system that analyzes chest X-Ray image data by using convolutional neural network is proposed to diagnose the common symptom,pneumonia. In this scheme,the image is preprocessed first,and then the features of the image of chest X-Ray are extracted by the VGG convolutional neural network and the improved VGG convolutional neural network respectively to obtain the accuracy rates of pneumonia detection of the two networks. The testing results indicate that the improved VGG convolutional neural network has better identification effect than the VGG convolutional neural network,and has faster convergence while training the demarcated dataset. The results of testing on public data sets show that the pneumonia detection system based on the improved VGG convolutional neural network has got the accuracy of 98.5%,that is,the accuracy is improved more than2% in comparison with the one based on the VGG convolutional neural network,which proves this system is feasible for pneumonia identification..