MoBMGAN: Modified GAN-Based Transfer Learning for Automatic Detection of COVID-19 Cases Using Chest X-ray Images

Novel COVID-19, a wide-reaching pandemic, has buckled the healthcare systems of the whole world. Suggested in this chapter is a computationally efficient automated detection scheme by utilizing modified Generative Adversarial Network (MGAN) based transfer learning (TL) for the successful detection of COVID-19 infection by using chest X-ray images. Initially, the MGAN model is used for the generation of a synthetic dataset with varied feature attributes which will serve as an excellent platform for the initial training of TL-based models (TLMs). MGAN is computationally efficient by utilizing lightweight network architecture. This model is free from the problem of instability by utilizing the weighted combination of content and structural loss for the training of the generator and discriminator block of the GAN model. Consequently, the MGAN model converges at a faster rate than the conventional GAN model and produces realistic generated samples. Afterward, various benchmark TLMs such as VGG19, ResNet50, InspectionV3, InspectionResnetV2, DenseNet121, DenseNet169, DenseNet201, and MobileNet are fine-tuned with the generated dataset and are utilized for the classification of COVID-19 infected X-ray images. COVID-19 detection by coupling MGAN and MobileNet model outperforms the other TLMs and provides an average accuracy, sensitivity, and F1 score of 99.71%, 93.48%, and 95.70%, respectively, at a lower computational burden. This chapter provides a computationally efficient automated detection scheme by utilizing modified Generative Adversarial Network (MGAN) based transfer learning (TL) for the successful detection of Corona Virus DIsease (COVID-19) infection by using chest X-ray images. Advanced artificial intelligence (AAI)-based methods utilizing various radiometric images provide an excellent breakthrough in detecting COVID-19-infected persons. Performance accuracy of deep learning (DL)-based classification methodology is severely influenced by the size of the training dataset required for the fine-tuning of parameters. The novel COVID-19 epidemic is recent. Performance behavior of each model has been recorded in terms of training and validation accuracy as well as training and validation loss. Transfer learning (TL) based approaches including different pre-trained deep learning models (DLMs) afford outstanding breakthroughs in the classification of COVID-19 cases from non-COVID cases by utilizing radiometric images such as computed tomography (CT) scans and