Mohammadpoor Mojtaba, Sheikhi Karizaki Mehran, Sheikhi Karizaki Mina
Electrical and Computer Department, University of Gonabad, Gonabad, Iran.
Department of Electrical and Computer Engineering, University of Birjand, Birjand, Iran.
PeerJ Comput Sci. 2021 Apr 1;7:e345. doi: 10.7717/peerj-cs.345. eCollection 2021.
COVID-19 pandemic imposed a lockdown situation to the world these past months. Researchers and scientists around the globe faced serious efforts from its detection to its treatment.
Pathogenic laboratory testing is the gold standard but it is time-consuming. Lung CT-scans and X-rays are other common methods applied by researchers to detect COVID-19 positive cases. In this paper, we propose a deep learning neural network-based model as an alternative fast screening method that can be used for detecting the COVID-19 cases by analyzing CT-scans.
Applying the proposed method on a publicly available dataset collected of positive and negative cases showed its ability on distinguishing them by analyzing each individual CT image. The effect of different parameters on the performance of the proposed model was studied and tabulated. By selecting random train and test images, the overall accuracy and ROC-AUC of the proposed model can easily exceed 95% and 90%, respectively, without any image pre-selecting or preprocessing.
在过去几个月里，新冠疫情使全球进入封锁状态。全球的研究人员和科学家在从新冠病毒的检测到治疗方面都付出了巨大努力。
病原学实验室检测是金标准，但耗时较长。肺部CT扫描和X光检查是研究人员用于检测新冠阳性病例的其他常用方法。在本文中，我们提出一种基于深度学习神经网络的模型，作为一种可替代的快速筛查方法，该方法可通过分析CT扫描来检测新冠病例。
将所提出的方法应用于一个公开可用的包含阳性和阴性病例的数据集，结果表明该方法能够通过分析每一幅CT图像来区分阳性和阴性病例。研究了不同参数对所提模型性能的影响并制成表格。通过随机选择训练图像和测试图像，所提模型的总体准确率和ROC-AUC分别可轻松超过95%和90%，无需任何图像预筛选或预处理。
