Mapping COVID-19 Artificial Intelligence (AI) Research in Medical Imaging: A Bibliometric Analysis of Datasets, Trends, and Clinical Challenges.
Authors
Affiliations (4)
Affiliations (4)
- College of Bioinformatics Science and Technology, Harbin Medical University, Harbin, CHN.
- School of Artificial Intelligence, Shenzhen University, Shenzhen, CHN.
- College of Computer Science and Artificial Intelligence, Wenzhou University, Wenzhou, CHN.
- College of Biochemistry and Molecular Biology, Harbin Medical University, Harbin, CHN.
Abstract
The rapid adoption of artificial intelligence (AI) for COVID-19 pandemic diagnosis has exposed critical gaps in medical imaging datasets. This Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA)-compliant bibliometric review of 450 PubMed studies (2020-2024) reveals that only 21.5% of the datasets remain clinically validated, while 55% are unavailable or repurposed from non-COVID-19 sources. We identified persistent issues, such as resolution heterogeneity and radiologist annotation scarcity, that undermine model reliability. Numerous convolutional neural network (CNN) architectures have been developed to enable fast and accurate automated diagnosis of COVID-19 using computed tomography (CT) or X-ray imaging. However, due to the urgency of the pandemic and the rapid demand for solutions, existing computer-aided diagnostic (CAD) systems face several critical limitations, such as imbalanced datasets, insufficient bias assessment in model training, and inconsistent quality control in image acquisition and preprocessing. In this bibliometric analysis, we provide an analysis of PubMed articles on COVID-19 imaging published between January 1, 2020, and November 1, 2024. The research included 1261 publications. VOSviewer was used to generate a visual map of the keyword networks and authors. The journal with the most publications was Elsevier, and the most used dataset was the COVID-19 Radiography Database from Kaggle.