Studies continue to pile up that show artificial intelligence can wield diagnostic prowess along with potential use in treatment planning and patient compliance monitoring. So far, however, AI’s net benefits for patients, clinicians, and health care institutions appear to remain largely theoretical. We write almost every week about new diagnostic applications for machine learning (ML) but ongoing implementations are rare. Last summer we referred to a study by a group of Toronto educators, researchers, and healthcare professionals that surveyed the issues and obstacles to implementing machine learning applications in medicine.
Another group of Canadians has found that barriers to ML implementations in medicine appear to be slowing the implementation of this new technology. Researchers from the University of Calgary’s Department of Clinical Neurosciences, Division of Neurosurgery published a study, “The impact of machine learning on patient care: A systematic review” in the March 2020, issue of Science Direct, Artificial Intelligence in Medicine.
The Calgary team reviewed the content of 1,909 publications from January 1, 2000, to May 1, 2018. The researchers found 386 articles that addressed machine learning applications with clinical problems. Of the included samples, 378 articles covered proof-of-concept studies while only eight involved implementations. It should not be a surprise that nearly two-thirds of the reviewed papers were published within the past four years. The recent development of new analytics tools have made machine learning projects more accessible.
The researchers concluded that any shift toward implementing machine learning applications with clinical medicine will coincide with increased usage of specialized tools designed to help care teams provide personalized patient care. As with any new technology — but especially one applied to medical situations — it will take time for these projects to evolve into useful systems that can move out of the lab and into the clinical setting.
