基于卷积神经网络的胸部X射线影像肺炎疾病分类研究

The Classification of Pneumonia Disease using Chest X-ray Imaging Based on Convolutional Neural Network

目的:自动检测肺炎以及区分2019冠状病毒病(COVID-19)和非COVID-19肺炎,旨在提高整体分类准确性.方法:数据集来自Kaggle存储库.使用编程环境MATLAB 2021a对所提出的模型进行开发和训练.该模型使用2913张胸部X线片图像(其中正常1005张,COVID-19900张,病毒性肺炎1 008张)进行训练,还使用从数据集中随机选择的一些未使用的胸部图像进行评估,并与现有深度学习方法相比较.结果:该模型在训练集上的平均准确率、召回率和精准率分别为0.989,0.983和0.984.此外,平均假阳性率和假阴性率分别为0.009和0.017.在验证集上,平均准确率、召回率和精准率分别为0.978、0.967和0.967.准确预测了未用于训练也未用于验证的图像60例(每类20例)中的58例.结论:利用卷积神经网络对胸部X线图像进行分类可以辅助放射科医师且能够减少他们之间可能由经验引起的图像解释的差异性.

Purpose:To automatically detect pneumonia and distinguish coronavirus disease 2019(COVID-19)and non COVID-19 pneumonia in order to improve the accuracy of overall classification.Methods:The dataset was from the Kaggle repository.The proposed model was developed and trained using the programming environment Matlab 2021a.The model was trained with 2913 chest X-ray images,of which 1005 were normal,900 were COVID-19 and 1008 were viral pneumonia.Some unused chest images randomly selected from the dataset were also evaluated and compared with existing depth learning methods.Results:The average accuracy,recall and precision of the model on the training set were 0.989,0.983 and 0.984,respectively.In addition,the average false positive rate and false negative rate were 0.009 and 0.017,respectively.In the validation set,the average accuracy,recall and precision were 0.978,0.967 and 0.967,respectively.Fifty-eight of the 60 images(20 of each class)that were not used for training and validation were accurately predicted.Conclusions:Using convolution neural network to classify chest X-ray images can assist radiologists and reduce the differences in image interpretation that may be caused by experience.