摘要
目的:构建基于胸片的深度学习模型检测新生儿肺炎,旨在提高新生儿肺炎的影像诊断水平及效率。方法:回顾性收集2018年1月至2021年9月暨南大学附属第一医院336幅新生儿胸片,其中新生儿肺炎176例,正常160例。随机将图像按8∶1∶1的比例分为训练集、测试集及验证集。利用ResNet50神经网络进行分类训练模型,然后进行验证。结果:验证组中深度学习模型诊断新生儿肺炎的曲线下面积AUC为0.9931。结论:深度学习模型能够准确诊断新生儿肺炎,但尚需大样本多中心进一步验证研究。
Objective:A chest radiograph-based deep learning model was developed to detect neonatal pneumonia,to improve the diagnostic accuracy;and efficiency of this disease.Methods:Three hundred and thirty-six chest radiographs of newborns were collected from the First Affiliated Hospital of Jinan University,including 176 cases of pneumonia and 160 normal cases.They were randomly divided into the training,validation,and test datasets at a ratio of 8∶1∶1.ResNet50 was used to train the classification model and then validate its performance.Results:In the validation dataset,the model achieved an area under the curve of 0.993.Conclusion:Deep learning model is accurate in the classification of neonatal pneumonia,however the further investigation is needed by using multicentre external validations.
作者
莫琦岚
张水兴
MO Qilan;ZHANG Shuixing(Department of Imaging,the First Affiliated Hospital,Jinan University,Guangzhou 510632,Guangdong,China)
出处
《暨南大学学报(自然科学与医学版)》
CAS
CSCD
北大核心
2022年第2期199-204,共6页
Journal of Jinan University(Natural Science & Medicine Edition)
基金
国家自然科学基金项目(81871323)。