摘要
目的为丰富儿科肺炎辅助诊断算法,提高医生分析儿科肺炎X线影像图片的效率和质量,提出一种改进的卷积神经网络模型。方法基于深度残差网络(ResNeXt-50),融合SE模块建立通道之间的关联,然后在模型构建过程中使用Leaky Re LU激活函数替代ReLU激活函数,并使用组归一化作为归一化方法,最后将预训练好的模型在Chest X-Ray数据集上进行训练测试,并以准确率(Accuracy)、召回率(Recall)以及精确率(Precision)作为评价指标。结果网络模型的识别准确率、精确率和召回率分别达到了91.19%、89.70%和91.39%。结论网络模型具有一定的实用性,能够更好地拟合肺炎图像数据集,能有效提升儿科肺炎图像分类的准确性,可作为临床上儿科肺炎的辅助诊断新方法。
Objective To propose an improved convolutional neural network model for enriching the auxiliary diagnosis algorithm of pediatric pneumonia and improving the efficiency and quality of X-ray image analysis of pediatric pneumonia.Methods Based on deep residual network(RESNext-50)and SE module was fused to establish the association between channels.Then,the Leaky ReLU activation function was used to replace the ReLU activation function in the model construction process,and group normalization was used as normalization method.Finally,the pre-trained model was trained and tested on Chest X-Ray dataset,and accuracy,recall and precision were used as evaluation indexes.Results The recognition accuracy,precision and recall of the network model reached 91.19%,89.70%and 91.39%,respectively.Conclusion The network model has a certain practicality,which can better fit the pneumonia image data set,effectively improve the accuracy of pediatric pneumonia image classification,and can be used as a new method for clinical diagnosis of pediatric pneumonia.
作者
张科
张春晓
ZHANG Ke;ZHANG Chunxiao(Medical Engineering Management Office,Shandong Provincial Hospital Affiliated to Shandong First Medical University(ShandongProvincial Hospital),Jinan Shandong 250021,China)
出处
《中国医疗设备》
2022年第9期42-46,56,共6页
China Medical Devices
关键词
图像分类
卷积神经网络
儿科肺炎
计算机辅助诊断
image classification
convolutional neural network
pediatric pneumonia
computer-aided diagnosis