Artificial intelligence (AI) is now the hottest keyword in the biomedical sphere with companies, including Philips, Lunit, and Vuno, launching new AI-based imaging aid to help doctors, especially radiologists, treat their patients better.
Reflecting the mounting interests, the Ministry of Food and Drug Safety recently approved nine AI-based medical device software and five related clinical trials. However, not all is bright as some skeptics fear that AI takes jobs away from doctors or turns out to be more of a problem than a solution.
Korea Biomedical Review met with two renowned radiologists on Friday during the AI leadership symposium titled “AI in Healthcare ‘Present and Future.’” The workshop was organized by The Korea Doctors Weekly, the sister paper of Korea Biomedical Review, to discuss on how AI has affected their work and the potential AI possess in the future.
The two radiologists are Professors Tim Leiner at Utrecht University Medical Center, and Peter B. Noël, director of CT Research at the University of Pennsylvania.
|Two radiologist experts explain the impact of AI in the field of the medical device at the Sheraton Seoul Palace Gangnam Hotel, on Friday. They are Professor Tim Leiner (left) at Utrecht University Medical Center, and Professor Peter B. Noël, director of CT Research at the University of Pennsylvania.|
KBR: How has AI medical imaging evolved in the medical field?
Leiner: Regarding the AI medical imaging development in the cardiovascular field, in the past couple of years, we’ve seen several companies come to the market with segmentation algorithms, and this is particularly interesting for cardiac magnetic resonance (MR).
This is because, in the past, doctors had to draw a lot of contours manually to calculate the ejection fraction.
However, the process has become a lot easier with automated algorithms, which can do the jobs on behalf of physicians. Such a help allows doctors to save 10 to 15 minutes when reading a case, and that adds up over a day or a week for physicians who read a lot of cases.
Noël: I think we are just in the point where we are seeing more and more applications coming into the clinics, while, at the same time, we have very high expectation for AI-based medical software from the medical or the clinical side as the interest for the area has been rapidly growing over the past decade.
I think we are at the tipping point where we’ll see the AI-based medical software used in the actual clinical setting.
KBR: Not all people welcome the rapid rise of artificial intelligence. Among the skeptics are people who are directly involved in the medical field, such as doctors. Why should we embrace AI in the medical field, and how far should we embrace it?
Leiner: I’m not one of those skeptics because I think that our job will become much more attractive when we develop the right set of tools.
People also often ask is if their jobs will be taken away because of AI, but I certainly don’t think that is the case.
By using AI-based software, we will be able to do a much better job in taking care of patients, and that’s the primary thing we should keep in mind as doctors work to serve their patients and give them proper medical care.
So, if algorithms can help us do this, I think we should embrace it and not be afraid that certain parts of what we do may change. I also believe that the change will probably be for the better because AI algorithms can take away some of the more tedious tasks so that doctors can spend the time saved into better patient care.
But we also have to implement AI in a relevant way and solve related problems.
Noël: When I first heard about the paradigm change of AI, I believed that it would change the way we work and reduce the number of people required. However, more and more I think about it, I don’t think this will be an issue because AI will help us improve how we diagnose patients. We still need doctors, and AI will only help them make a better decision.
With AI, we can actually ensure that we offer every single patient the same type of service while helping a younger generation of doctors educationally.
So I think no physicians have to worry about their jobs as it will only improve our workflow.
KBR: Another problem mentioned by practitioners here is that AI only provides the final diagnosis without the process of reaching such conclusions known as the black box issue. Such a process might be okay for experienced physicians, but it may hinder the learning process of less experienced doctors. This may become a problem because many believe that even if AI develops, it will be up to doctors to make the final decision. Any thoughts?
Leiner: That’s a good point and a concern for everyone, but whether or not this is an issue depends on the task at hand.
As a cardiovascular radiologist, I don’t care if it’s a black box algorithm that does the segmentation of the left ventricle as long as it does it correctly. However, if we talk about prognosis and an AI algorithm tells a patient that he has five more years to live than we want to know why and what we can do about it.
I think we do need an explanation on which type of information the AI algorithm based its decision. But, even in this area, companies are working on providing such an explanation, and I believe that the prognosis field will grow in the coming years.
So whether or not this is an issue depends on the exact question that a physician is trying to answer.
Noël: I think this is one of the points that we, as a community, have to learn how to handle as I don’t think it is sufficient for an AI algorithm to tell us if a patient has a specific disease. I believe it is essential to see how the algorithm makes its decision, and this is important not only for the younger generation of radiologists but also for experienced radiologists because how can we base our approval of the AI algorithm if we don’t understand how a certain decision has been made.
KBR: AI in healthcare is quite a new concept, and regulatory officials have recently been rolling out new regulations in this area. In your opinion, how should regulatory agencies approach making new regulations in the field?
Leiner: I think that the most important thing is that they take it to account that these algorithms can learn over time. This is a big difference when compared to the prior generation of computer-aided diagnosis (CAD) software that had a fixed set of parameters and couldn’t really learn.
Such software would always make the same mistakes, and the mistakes would not go away, but now, when we work with an AI algorithm, it has the potential to learn from its mistakes and improve over time.
This is something the regulatory agencies have to start to think about.
But I think that global officials are open to such an idea and they are trying to draw up rules and regulations that can deal with the issue at hand.
I believe that we can expect algorithms to undergo much more frequent updates compared to CAD programs that we had in the past.
Noël: I think it is crucial to think about how to implement the AI-based medical device software the right way and not the fast way. I think it is also in our nature that we believe that AI will come and affect our lives soon, but we know from experience that such implementations take time to get in place.
KBR: What do you think the future of AI is in the healthcare field?
Leiner: I think the future is very bright, as AI has a tremendous potential to improve the whole healthcare process. Every patient has so many data points, such as the electronic healthcare record, laboratory, pathology, and radiology data, that it is almost impossible for a human to extract all the information out of the data sets, and AI can really help us do a better job in this area. This will help us provide really personalized treatment for patients.
Noël: I think it will be an essential part of our healthcare system and play specific roles in the hospital.
<© Korea Biomedical Review, All rights reserved.>