Trends in artificial intelligence for cardiovascular research: a topic modelling study using BERTopic

Abstract

Abstract Background Research on Artificial Intelligence (AI) implementation in medicine, including cardiology, has grown rapidly in recent years. Understanding dynamic changes in the topic will help researchers and practitioners navigate the future of research and healthcare. Purpose The purpose of this study is to analyse the use of AI in Cardiovascular research. By utilising topic modelling, we could identify dominant topics and trend changes over a decade, including the most commonly used AI methods and how they have changed over time. Methods We analysed 12,431 research abstracts published from 2015 to 2024 using keyword combinations related to "AI" and "Cardiology" from PubMed. For the topic extraction, we used BERTopic, an advanced topic modelling technique paired with SPECTER2 embedding adapter, a model pre-trained specifically for scientific texts, to ensure more accurate, semantically richer, and contextually relevant topic clustering compared to traditional non-specific domain models. Results Our modelling results show an upward trend in several clusters of cardiology themes, such as ‘electrocardiogram signal classification’, ‘cardiac image segmentation’, and ‘cardiovascular disease risk prediction modelling’. Among the top ten themes, some research themes appear to have stagnated compared to others, such as ‘chest CT classification’. Even research on the impact of COVID-19 on the cardiovascular system, which had been growing rapidly until 2021, experienced a significant decline in subsequent years. Through the BERTopic approach, we also observed the evolution of keywords emerging in each topic cluster from year to year. While some main topics have consistent research topics, with the advancement of AI technology, exploration toward complex topics such as cine-cardiac imaging began to dominate in 2021. This analysis also shows that the most frequently used algorithms are Random Forest, Convolutional Neural Networks (CNN), and Logistic Regression. Interestingly, with the popularity of AI paradigms such as ChatGPT and video-based processing, there has been a significant increase in the application of appropriate AI architectures, reflecting a new direction in integrating modern AI into cardiovascular research. Conclusion This study concludes that the AI research landscape in cardiology has a dual evolutionary trend, with fundamental AI models becoming the preferred choice for strengthening established research themes. On the other hand, rapid evolution and modern AI allow new topics to emerge and more complex problems to be solved. This divergence shows complementary directions. One side focuses on refining proven methods, while the other focuses on driving innovation through modern architecture and richer data modalities, which can serve as a compass in selecting the right topics and AI utilisation in future research.Cardiovascular themes over timeThemes and AI architectures

Ähnliche Arbeiten