​Physicians might find a superhuman assistant by putting computer vision to work in a clinical setting, according to an April 5, 2018 perspective in The New England Journal of Medicine. A type of artificial intelligence more popularly known as one of the technologies that makes self-driving vehicles possible, these machines have the ability to see and understand visual input and then learn from it by amassing data and analyzing the patterns within, rather than relying on instruction specified by a human programmer. Computer vision could potentially spot a healthcare provider's bedside errors such as failing to reset bedrails or restraints, or it could be used to screen medical images for abnormalities, according to the article. Other possibilities include reminding a surgeon of a missed step during a procedure, or notifying a nurse that a patient may be trying to remove an endotracheal tube. One California hospital has used the technology to observe behavior concerning hand-hygiene compliance, using depth and thermal sensors in order to allay privacy concerns. At this facility, the machine has so far been 84.6% accurate in detecting hand hygiene events. But, the author says, artificial intelligence in the clinical setting has induced skepticism after some early uses of the technology performed below expectations. Instances of poor data quality, a failure to work well in the usual clinical workflow, and the difficulty of explaining the complex computing it performs are all roadblocks to universal implementation or acceptance.

HRC Recommends: Risk managers should remain alert to emerging technologies that can affect patient care and safety. As they do with any new technology, healthcare organizations must evaluate the technology's risks and benefits to patient care. If a technology such as machine learning–decision support systems (ML-DSS) becomes ready for adoption, healthcare organizations must evaluate the ease of integrating the technology into clinicians' workflow—as well as the possibility that the new technology can introduce errors in patient care—before implementing it.