HEAVY-SARS: Harnessing Explainable AI and Generative Models for Visual Synthetic Data Generation of SARS-CoV-2

Technological advancements have facilitated the creation of extensive real-world data, although deriving valuable insights from these substantial datasets continues to be labour-intensive and expensive. The production of synthetic data with Generative Adversarial Networks (GANs) provides a viable approach. This study integrates the network architecture of DCGAN with the parameter optimisation technique of WGAN to produce synthetic CT scan data. We assess the effectiveness of deep learning models trained on synthetic data by juxtaposing them with those trained on actual data, utilizing XAI LIME. The experimental findings illustrate the efficacy of the suggested model in producing high-quality synthetic data. The model demonstrates similar performance on both actual and synthetic datasets, achieving accuracy values of 0.9818 for real data and 0.9745 for synthetic data under identical conditions. These findings highlight the applicability of our technique in real-world contexts, especially in situations with restricted access to extensive datasets.