Artificial intelligence in COVID-19 research: A comprehensive survey of innovations, challenges, and future directions

Abstract

The COVID-19 pandemic has accelerated the use of AI and ML in healthcare, improving diagnosis, treatment, and resource allocation. This survey examines the AI applications in disease detection, differential diagnosis, and post-COVID complication analysis. Our findings show that 53% of the reviewed studies focus on COVID-19 detection, while only 14% address post-COVID complications. This reveals a gap in long-term patient monitoring. Convolutional Neural Networks (CNNs) are the most frequently used models, appearing in 23% of reviewed studies as standalone architectures and even more often in hybrid models. Meanwhile, transformers and multimodal models remain underutilized. Each appears in only 4% of the studies, limiting the integration of diverse data sources, such as imaging, audio, and lab results. Federated learning, a privacy-preserving AI approach, appears in 9% of studies. However, it is still less common than centralized models. This restricts secure and collaborative AI development. Despite these progress, challenges such as data bias, limited model generalization, and ethical concerns persist. Advanced methods including transformer models and knowledge distillation offer potential solutions for improving computational efficiency. To strengthen AI-driven healthcare, this survey highlights three key needs: (1) broader adoption of multimodal AI, (2) development of computationally efficient and interpretable AI models, and (3) increased use of federated learning to support privacy-preserving AI training. By synthesizing insights from various studies, this paper provides a comprehensive evaluation of AI innovations in COVID-19 research and outlines key directions for future advancements in ethical and scalable AI-driven healthcare.