Artificial intelligence seems to be everywhere nowadays, and, depending on who you ask, it’s either the dawn of a new age or the end of civilization. AI’s potential worries some physicians who might find themselves not needed, or at least not needed as much. They also are concerned about whether they can stay on top of the developments in AI and keep up with the best ways to harness its power.
What is undeniable is that AI is here, and its uses are growing faster than ever. According to the American College of Radiology, 30% of radiologists are already using AI in their clinical practices.
“It is a tool that is helping me be a better radiologist and physician,” said Dr. Reggie Munden, a radiologist who specializes in reading images of the lungs and chest at the Medical University of South Carolina.
This seems to be the prevailing opinion – that AI is a tool that helps physicians do their jobs better. Programs based on AI can quickly examine hundreds of images and call a doctor’s attention to possible problems. Dr. Munden sees AI as a way to increase his efficiency and effectiveness.
“The efficiency gain is material,” said Chad Wiggins, system administrative officer, imaging service line, with MUSC.
As interest in AI has grown, so have the ways medical professionals have found to use it. AI in radiology is classifying brain tumors, diagnosing ALS, detecting Alzheimer’s disease and more. AI programs can read images faster without wavering focus or concentration, and in some cases, it might catch something a radiologist might not have noticed.
Dr. Munden added, however, that “It’s critically important to know that it is not replacing us.”
What it does do is eliminate some of the “clutter” that radiologists might have to deal with. An MRI can produce hundreds of images, but most will have nothing to do with the reason a patient is getting the MRI. An AI program can quickly eliminate these images, leaving the pertinent ones to be reviewed.
One reason radiologists will continue to be needed is that AI programs are designed to do specific tasks. Human oversight is still needed to review the images an AI program flags to verify whether there is an issue. While the programs are getting better at this, they are not capable of coordinating input from other fields or developing a comprehensive medical plan based on a variety of data.
While AI programs have improved efficiency and speed, they can still make mistakes. As with humans, it all comes down to training.
AI programs have been fed millions of images before they are even brought online, but, even so, they have what is called a training bias. Their “experience” is based on a database of images that are run through algorithms, which have their own programming bias. A new problem that isn’t in that database could confound the program, which is why an experienced human radiologist acts as a second opinion.
“It’s only as good as the data we feed into it, only as good as the information we feed into it,” said Dr. Morgan McBee, a pediatric radiologist and associate professor of radiology and pediatrics with MUSC.
AI programs are only expected to improve, which will not only enhance the way they fulfill current applications but also open new possibilities.
“Right now, we are just grabbing the low-hanging fruit,” said Dr. McBee. “It’s still a pretty new idea.”
There are hundreds of radiology products on the market that have been approved by the Food and Drug Administration. Some have been trained to perform different applications. They have different algorithms, different training and different databases of images they were trained on – all of which can lead to varying results.
This also means medical staffs need to review the most promising programs that are said to meet their needs and see if they live up to the hype. For instance, Wiggins said that as AI effectiveness continues to be validated, it could be used in emergency departments to increase their responsiveness.
One final reason that radiologists and other medical professionals will continue to be needed is that patients need that human touch. Programs can’t empathize with patients, calm their fears or show a genuine interest in their life.
As a tool, AI is a breakthrough – but for a physician, AI is still just a tool.
By James Rada Jr.
