Frenzied reporting of the potential impacts of artificial intelligence (AI) would lead any radiologist to fear a robot revolution is just around the corner. However, the reality is much more mundane, according to Claudio Silvestrin, head of the AI Centre of Excellence at Unilabs.
While AI undoubtedly has the potential to help transform radiology in the future, as it stands “there is no widespread adoption,” he says.
The Centre of Excellence is evaluating promising AI technologies, but at present the state of the art is hardly ground breaking.
Instead, when AI starts to enter radiology practice it is likely to be in the form of unobtrusive productivity- and quality-enhancing tools, such as applications that highlight potential areas of concern in CT scans, rather than big-bang innovations.
Narrow AI applications are hitting the market. But Silvestrin says Dr Bradley Erickson of the Mayo Clinic in Rochester, Minnesota, was probably right last year in predicting it would take up to three years for deep-learning algorithms to create full preliminary reports for mammography.
They would still need to be checked by humans, however.
Based on Erickson’s estimations, it could be a decade before AI is able to create full preliminary reports on CT scans of the head, chest, abdomen and pelvis, MR images of the head, knee and shoulder, or ultrasound views of the liver, thyroid and carotids.
The point when deep-learning algorithms can produce full reports for most diagnostic imaging studies, meanwhile, could be up to 20 years away. That doesn’t mean AI won’t see progress in the near future, though.
On the contrary, units such as the AI Centre of Excellence are likely to be very busy. “There are thousands of potential applications in radiology alone,” Silvestrin says; the issue, for now, is finding the ones that make most sense.
- To find out more about the AI Centre of Excellence, contact Claudio Silvestrin