Technological developments have left an enormous imprint on major health care factors, such as cost, quality, and access by patients. A current topic of significant importance is the realm of activity known as artificial intelligence (AI). The various uses of AI in medicine have been expanding rapidly in many areas, including in the: analysis of medical images, detection of drug interactions, identification of high-risk patients, and coding of medical notes. Several such uses are topics in the “AI in Medicine” review article series that had its debut in the March 30, 2023 issue of the New England Journal of Medicine. An aim of the series is to cover progress, pitfalls, promise, and promulgation at the interface of AI and medicine. As a further commitment, a new journal, NEJM AI, will be launched in 2024 to provide a forum for high-quality evidence and resource sharing for medical AI, along with informed discussions of its potential and limitations.
As a consequence of a substantial investment of money and intellectual effort, computer reading of electrocardiograms (ECGs) and white-cell differential counts; analysis of retinal photographs and cutaneous lesions; and other image-processing tasks has become a reality. Many of these machine-learning–aided tasks have been largely accepted and incorporated into the everyday practice of medicine while the use of AI and machine-learning in medicine has expanded beyond the reading of medical images. AI and machine-learning programs have entered medicine in ways that include, but not limited to, helping to identify outbreaks of infectious diseases that may have an influence on public health; combining clinical, genetic, and many other laboratory outputs to identify rare and common conditions that might otherwise have escaped detection; and aiding in hospital business operations.
As noted in the NEJM, the use of AI and machine-learning already has become accepted medical practice in the interpretation of some types of medical images, such as plain radiographs, computed tomographic (CT) and magnetic resonance imaging (MRI) scans, and skin images. For these applications, AI and machine-learning have been shown to help health care providers by flagging aspects of images that deviate from the norm. A key question becomes what is the norm? This simple query reveals one of the weaknesses of the use of AI and machine-learning in medicine as it is largely applied today.
Key concerns requiring a much deeper understanding include how bias in the way AI and machine-learning algorithms were “taught” influence how they function when applied in the real world? How can human values be interjected into AI and machine-learning algorithms so that the results obtained reflect the real problems faced by health professionals? What issues must regulators address to ensure that AI and machine-learning applications perform as advertised in multiple-use settings? How should classic approaches in statistical inference be modified, if at all, for interventions that rely on AI and machine-learning? These problems are among the many that must be confronted. The “AI in Medicine” series can be expected to address these kinds of matters.