Hosseinzadeh Hamidreza
Department of Electrical Engineering, North Tehran Branch, Islamic Azad University, Tehran, Iran.
Biomed Signal Process Control. 2022 May;75:103595. doi: 10.1016/j.bspc.2022.103595. Epub 2022 Feb 23.
COVID-19 is a pandemic infectious disease which has influenced the life and health of many communities since December 2019. Due to the rapid worldwide spread of this highly contagious disease, making its early detection with high accuracy important for breaking the chain of transition. X-ray images of COVID-19 patients, reveal specific abnormalities associated with this disease.
In this study, a multi-view feature learning method for detecting COVID-19 based on chest X-ray images is presented. This method provides a framework for exploiting the multiple types of deep features, which is able to preserve both the correlative and the complementary information, and achieve accurate detection at the classification phase. Deep features are extracted using pre-trained deep CNN models of AlexNet, GoogleNet, ResNet50, SqueezeNet, and VGG19. The learned feature representation of X-ray images are then classified using ELM.
The experiments show that our method achieves accuracy scores of 100%, 99.82%, and 99.82% in detecting three classes of COVID-19, normal, and pneumonia, respectively. The sensitivities of three classes are 100%, 100%, and 99.45%, respectively. The specificities of three classes are 100%, 99.73%, and 100%, respectively. The precision values of three classes are 100%, 99.45%, and 100%, respectively. The F-scores of three classes are 100%, 99.73%, and 99.72%, respectively. The overall accuracy score of our method is 99.82%.
The results demonstrate the effectiveness of our method in detecting COVID-19 cases and can therefore assist experts in early diagnosis based on X-ray images.
自2019年12月以来，新型冠状病毒肺炎（COVID-19）是一种大流行性传染病，已影响到许多社区的生活和健康。由于这种高传染性疾病在全球迅速传播，因此高精度地早期检测对于打破传播链至关重要。COVID-19患者的X线图像显示出与该疾病相关的特定异常。
在本研究中，提出了一种基于胸部X线图像检测COVID-19的多视图特征学习方法。该方法提供了一个利用多种深度特征的框架，能够保留相关信息和互补信息，并在分类阶段实现准确检测。使用预训练的AlexNet、GoogleNet、ResNet50、SqueezeNet和VGG19深度卷积神经网络（CNN）模型提取深度特征。然后使用极限学习机（ELM）对学习到的X线图像特征表示进行分类。
实验表明，我们的方法在检测COVID-19、正常和肺炎三类时的准确率分别达到100%、99.82%和99.82%。三类的灵敏度分别为100%、100%和99.45%。三类的特异性分别为100%、99.73%和100%。三类的精确率分别为100%、99.45%和100%。三类的F值分别为100%、99.73%和99.72%。我们方法的总体准确率为99.82%。
结果证明了我们的方法在检测COVID-19病例方面的有效性，因此可以帮助专家基于X线图像进行早期诊断。
