基于深度学习的COVID-19智能诊断系统  

Intelligent Diagnosis System for COVID-19 Based on Deep Learning

作　　者：贾楠 李燕[2] 郭静霞[1] 徐立[1] 白金牛[1] JIA Nan;Li Yan;GUO Jingxia;XU Li;BAI Jinniu(Baotou Medical College,Inner Mongolia University of Science and Technology,Baotou 014040,China;Baotou Central Hospital,Baotou 014040,China)

机构地区：[1]内蒙古科技大学包头医学院计算机科学与技术学院,内蒙古包头014040 [2]包头市中心医院,内蒙古包头014040

出　　处：《计算机测量与控制》2023年第4期96-103,共8页Computer Measurement &Control

基　　金：内蒙古自治区高等学校科学研究项目(NJZY22050);包头医学院研究基金项目(BYJJ-KCRH202206)。

摘　　要：针对2019年12月在全球爆发的新冠肺炎,传统的RT-PCR检测易出现假阴性或弱阳性并且检测时间长等问题,设计了一种基于深度学习对胸部X光片辅助诊断新冠肺炎的方法;采用UNet分割模型实现了肺部ROI区域的自动分割,对分割后的影像应用TrivialAugment数据增强策略,通过MBCA-COVIDNET模型实现胸部X光片三分类(新冠肺炎、其它肺炎、正常)任务,该模型以MobileNetV2作为骨干网络,并在其中加入坐标注意力机制(CA);利用Hugging Face和Flask开源软件搭建了COVID-19智能诊断系统;实验结果表明MBCA-COVIDNET模型在COVID-QU-Ex Dataset测试集上取得了高达97.98%的准确率,而模型的参数量和MACs仅有2.23 M和0.33 G,该智能诊断系统能够很好地辅助医生进行基于胸片的COVID-19诊断,提升诊断的准确率以及诊断效率。Aimed at the global outbreak of COVID-19 in December 2019,traditional RT-PCR tests are prone to the questions of false negative,weak positive results and long detection,A deep learning-based method for assisting in the diagnosis of COVID-19 on chest X-ray was proposed.The UNet segmentation model is used to accomplish the automatic segmentation of the lung ROI area,the segmented image is enhanced by the Trivial Augment data enhancement strategy,and the X-ray images of the chest are classified into three categories(normal,COVID-19,and other pneumonia).The model adds coordinate attention mechanism(CA)to MobileNetV2 as the backbone network;the open-source software such as Hugging Face and Flask is adopted to construct the COVID-19 intelligent diagnosis system;The experimental results show that the MBCA-COVIDNET model reaches the accuracy of 97.98%on the COVID-QU-Ex Dataset test set,even though the model's parameters and MACs are 2.23 M and 0.33 G respectively.The intelligent diagnostic system can help doctors diagnose COVID-19 based on the chest X Ray(CXR),and improve the accuracy and efficiency of the diagnosis.

关 键 词：胸片 新冠肺炎 深度学习 分类模型 坐标注意力机制 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]