A novel vision transformer model for rumor prediction in COVID-19 data CT images

Abstract

The convergence of healthcare and deep learning has engendered transformative solutions for myriad medical challenges. Amid the COVID-19 pandemic, innovative strategies are imperative to mitigate the propagation of misinformation and myths, which can exacerbate the crisis. This study embarks on a pioneering research quest, harnessing advanced deep learning methodologies, including the novel Vision Transformer (ViT) model and state-of-the-art (SOTA) models, to predict and quell the dissemination of rumors within the COVID-19 milieu. By synergizing the capabilities of Vision Transformers (ViTs) with cutting-edge SOTA models, the proposed approach strives to elevate the precision of information disseminated through traditional and digital media platforms, thereby cultivating informed decision-making and public awareness. Central to this inquiry is the development of a bespoke vision transformer architecture, adeptly tailored to scrutinize CT images associated with COVID-19 cases. This model adeptly captures intricate patterns, anomalies, and features within the images, facilitating precise virus detection. Extending beyond conventional methodologies, the model adroitly harnesses the scalability and hierarchical learning intrinsic to deep learning frameworks. It delves into spatial relationships and finer intricacies within CT scans. An extensive dataset of COVID-19-related CT images, encompassing diverse instances, stages, and severities, is meticulously curated to fully exploit the innovative potential of the vision transformer model. Thorough training, validation, and testing refine the model’s predictive prowess. Techniques like data augmentation and transfer learning bolster generalization and adaptability for real-world scenarios. The efficacy of this research is gauged through comprehensive assessments, encompassing sensitivity, specificity, and prediction accuracy. Comparative analyses against existing methods underscore the superior performance of the novel model, highlighting its transformative influence on predicting and mitigating rumor propagation during the COVID-19 pandemic. Enhanced interpretability sheds light on the decision-making process, augmenting the model’s utility within real-world decision support systems. By harnessing the transformative capabilities of vision transformers and synergizing them with advanced SOTA models, this study offers a robust solution to counter the dissemination of misinformation during the pandemic. The model’s proficiency in discerning intricate patterns in COVID-19-related CT scans signifies a pivotal leap toward combating the infodemic. This endeavor culminates in more precise public health communication and judicious decision-making, ushering in a new era of leveraging cutting-edge deep learning for societal well-being amidst the challenges posed by the COVID-19 era.