AI efforts are still early, ‘not at the solution level’
While there has been a lot of publicity about the potential of AI in radiology to recognize images and, when paired with algorithms, assist in the diagnosis of diseases, this work remains in the research realm.
“We are not at the solution level yet,” says Paul Chang, MD, professor and vice chairman of radiology informatics at the University of Chicago School of Medicine. “It is a long way between initial promising results in the lab and true clinically validated solutions.”
And despite the implementations of machine-learning assisted algorithms for specific use cases in clinical operations, there are many stumbling blocks to widespread adoption of machine learning and other AI technologies. There are competing priorities for spending IT budgets and myriad technical issues, such as the lack of reliable large data sets, sophisticated data management or workflows that can incorporate new insights.
Nonetheless, health system executives believe the knowledge generated from AI tools will help their organizations remain financially viable as they assume financial risk for clinical outcomes. Machine learning, for example, can help derive predictive and prescriptive information from data, which then can be embedded into electronic health records, so that the insights are readily available to clinicians as they interact with patients.
“The real bottom-line benefit of this technology is understanding the business of medicine, so I can actually become more efficient, and reduce error and reduce variability,” Chang explains.
Using AI to draw conclusions from medical information, such as that contained in electronic medical records— say, as a clinician would—is more complex and much farther out on the horizon, says Luciano Prevedello, MD, division chief in medical imaging informatics at Ohio State University Wexner Medical Center.
“I can develop an algorithm that helps me detect whether there is blood in the brain or not—that is very doable,” he says. “But to get to the reason this patient is developing the hemorrhage? That is a different question and a different level of complexity.”