감염병 위기에서 AI는 어떻게 사용되었는가: COVID-19 연구동향 분석

Abstract

Background: The COVID-19 pandemic prompted a surge in artificial intelligence (AI) research with health- related applications, including diagnosis, forecasting, and policy evaluation. While many reviews have summarized model performance, few have examined the structural relationships between research aims, data types, and algorithm selection. Objectives: This study presents a narrative review of AI-based COVID-19 research, focusing on how algorithm choices evolved across different functional goals—clinical diagnosis and treatment, infection forecasting, and public health or policy response—and how these choices were shaped by data characteristics and different phases of the pandemic. Methods: We reviewed 108 peer-reviewed English-language studies published between 2020 and 2024. Each study was categorized by functional objective and analyzed in terms of AI methods used, data types and sources, geographic focus, and modeling strategies. Both quantitative trends and qualitative insights were synthesized. Results: Convolutional neural networks, support vector machines, and random forests were frequently used and showed broad applicability. Algorithm selection aligned closely with data types: deep learning was dominant in image-based tasks, while structured data often employed tree-based or logistic models. Over time, reliance on public case data declined, while the use of clinical and policy datasets increased. Supervised learning approaches remained dominant, although unsupervised methods were used for sentiment analysis and clustering. Modeling patterns varied significantly by research purpose, reflecting the structural match between methodological design and data context. Conclusions: AI use in COVID-19 research evolved with changing data environments and research needs. Algorithm choice reflected not only technical capacity, but also alignment with functional objectives and data structures. This narrative review provides guidance for the future development of AI-based tools in public health emergencies.