Improving effectiveness of different deep learning-based models for detecting COVID-19 from computed tomography (CT) images

Computerized Tomography (CT) has a prognostic role in the early diagnosis of COVID-19 due to it gives both fast and accurate results. This is very important to help decision making of clinicians for quick isolation and appropriate patient treatment. In this study, we combine methods such as segmentation, data augmentation and the generative adversarial network (GAN) to improve the effectiveness of learning models. We obtain the best performance with 99% accuracy for lung segmentation. Using the above improvements we get the highest rates in terms of accuracy (99.8%), precision (99.8%), recall (99.8%), f1-score (99.8%) and roc acu (99.9979%) with deep learning methods in this paper. Also we compare popular deep learning-based frameworks such as VGG16, VGG19, Xception, ResNet50, ResNet50V2, DenseNet121, DenseNet169, InceptionV3 and InceptionResNetV2 for automatic COVID-19 classification. The DenseNet169 amongst deep convolutional neural networks achieves the best performance with 99.8% accuracy. The second-best learner is InceptionResNetV2 with accuracy of 99.65%. The third-best learner is Xception and InceptionV3 with accuracy of 99.60%.