Artificial Intelligence in Biomedical Sciences: A Scoping Review

Abstract

Background: Artificial intelligence (AI) is increasingly playing important roles in healthcare diagnosis, treatment, monitoring, and prevention of diseases. Despite this widespread implementation of AI in biomedical sciences, it has yet to be characterized. Aim: The aim of this scoping review is to explore AI in biomedical sciences. Specific objectives are to synthesize six scopes addressing the characteristics of AI in biomedical sciences and to provide in-depth understanding of its relevance to education. Methods: This scoping review has been developed according to Arksey and O’Malley frameworks. PubMed, Embase, and Web of Science databases were searched using broad search terms without restrictions. Citations were imported into EndNote for screening and extraction. Data were categorized and synthesized to define six scopes discussing AI in biomedical sciences. Results: A total of 2,249 articles were retrieved for screening and extraction, and 192 articles were included in this review. Six scopes were synthesized from the extracted data: Scope (1): AI in biomedical sciences by decade, highlighting the increasing number of publications on AI in biomedical sciences. Scope (2): AI in biomedical sciences by region, showing that publications on AI in biomedical sciences mainly originate from high-income countries, particularly the USA. Scope (3): AI in biomedical sciences by model, identifying machine learning as the most frequently reported model. Scope (4): AI in biomedical sciences by discipline, with microbiology the discipline most commonly associated with AI in biomedical sciences. Scope (5): AI in biomedical sciences education, which was limited to only six studies, indicating a gap in research on the educational application of AI in biomedical sciences. Scope (6): Opportunities and limitations of AI in biomedical sciences, where major reported opportunities include efficiency, accuracy, universal applicability, and real-world application. Limitations include; model complexity, limited applicability, and algorithm robustness. Conclusion: AI has generally been under characterized in the biomedical sciences due to variability in AI models, disciplines, and perspectives of applicability.