by John R. Fischer
, Senior Reporter | March 02, 2018
From the January/February issue of HealthCare Business News magazine
Dr. Michael Recht, Lewis Marx Professor and Chair of the Department of Radiology at NYU Langone Health, spoke with HealthCare Business News about why he believes artificial intelligence is poised to have a dramatic impact on the entire workflow of radiologists, from scheduling an examination to the exam protocol to how an image is interpreted and how that interpretation is communicated to the patient.
“There are a number of articles that show pattern recognition algorithms doing as well as, and in some cases better than, radiologists on specific tasks,” says Recht. He and a few colleagues recently highlighted some of that research in an article published last year in the Journal of the American College of Radiology.
Ultimately, he believes these innovations will usher in a brand new way of managing a radiology department, understanding which patients will be no-shows, how to schedule so you can account for no-shows and add-ons and most efficiently utilize your equipment.
The right combination of workflow, clinical equipment and technology can improve patient care, clinical outcomes and your bottom line. Click above to learn more.
They will also change the way images themselves are interpreted, but there is still plenty of progress that needs to be made on that front. One challenge is that all of these emerging algorithms are what Recht calls narrow AI. They answer a specific question. For example, is there a pneumothorax or not a pneumothorax? Is there a nodule or is there not a nodule?
“The problem is,” explains Recht, “when we look at an image, we don’t look at a specific question.”
In order for machine learning to evaluate an image from the ground up, the way a radiologist does, would require general AI, and that level of sophistication may still be a good ways down the road.
For Recht and his colleagues, one particularly interesting and promising area of artificial intelligence is with image reconstruction, particularly for MR and CT.
“With MR, the huge advantage with machine learning reconstruction, and we have some early results to show, that you can undersample images and significantly speed up your acquisitions,” says Recht, who believes these algorithms could increase the speed of an MR scan by four to six types the current speed.
“We’re talking about potentially using MR in far more cases than we do now, and MR becoming less expensive than it is now,” he says.
With CT, undersampling means significantly reducing the radiation dose to the patient and potentially disrupting the way we think about imaging altogether.
“One of the things we’ve talked about in certain use cases is eliminating radiography,” he says. “If we can do a CT or an MR at a much lower price, or much lower radiation, or much faster than we currently do, we might not need X-ray anymore because we know we get far more information on these cross-sectional modalities than we do on radiography, so we’re very excited.”
Recht believes that innovations with undersampling and image reconstruction could also lead to more simplified imaging machines, like an MR that does not require all the magnetic homogeneity that currently defines the technology.
“I can picture a future where all of these AI tools present information to us in a much more efficient manner, putting the images in order and presenting them in a better way and using pattern recognition algorithms so that our productivity as radiologists could be significantly increased and we could read more and be much more integrated into the clinical care teams, working with the referring physicians and understanding the information, understanding what you do in terms of what imaging should be done, what follow-up should be done,” says Recht.
“It’s really changing our role a little bit and getting us much more integrated than a lot of radiologists are today."