Introduction
As of February 2023, the US Food & Drug Administration (FDA) has
approved 521 medical applications that utilize Artificial Intelligence
(AI) and Machine Learning (ML). Most of these (75%) are in radiology,
followed by cardiology, hematology and neurology. Similar trends are
observed in Conformité Européene (CE)-marked medical devices
incorporating AI within the European Union. Currently, no registered AI
and ML-based applications are being utilized in the field of allergy and
immunology. One can therefore question if this field is missing out on
new research opportunities and clinical applications either because of
insufficient access to AI applications or a lack of awareness of
potential applications. However, given the rapid pace of technological
advances, it can be anticipated that AI and ML algorithms will be
increasingly applied in allergy and immunology soon.
Over the past decade, medicine has witnessed an exponential growth of
interest in AI and the yearly number of scientific articles on AI has
increased tenfold since 2012. This trend is fueled by the explosion of
(bio)medical data, including multi-omics, image data, and digital
electronic health records (EHRs), along with advancements in computing
power. These developments have paved the way for advanced analytical
approaches to address new research questions on large-scale datasets.
Traditional analytical techniques are no longer adequate to handle such
data complexity, volume, and structure. The introduction of accessible
software and methodological advancements in AI have further promoted the
use of AI in the (bio-)medical field. Most importantly, ML and AI can
identify complex patterns in vast amounts of data, such as images, text,
or audio and deliver superior predictive power, often surpassing
traditional statistical methods.
This review provides a fundamental understanding of ML and AI’s core
concepts. A framework is presented to structure the broad umbrella term
AI, and an overview of several state-of-the-art applications of AI in
medicine and allergy and immunology, specifically, is provided. The
focus is on applications that preferably adhere to any, and ideally
multiple, of the following conditions: (1) are externally or
prospectively validated, (2) demonstrate a positive effect on clinically
relevant patient outcomes, (3) FDA and/or CE approval, (4) outperform
traditional methods, and (5) answer research questions where traditional
analytical techniques fail. Additionally, we critically discuss the
limitations and open challenges of AI applications and share an outlook
on good practices of AI and ML in allergy and immunology.