The future of radiology AI (Artificial Intelligence) continues to grow as machine learning expands its capabilities into new industries. At ECR 2018, artificial intelligence was a theme on opening day as Professor Wiro Niessen from Biomedical Imaging Group Rotterdam, the Netherlands, addressed AI in the press conference. Declaring AI an enormous opportunity for the field, Niessen confirmed his standpoint that the radiological community must work together with those in machine learning, in order to ensure that AI has a positive impact on medicine.
Although AI will undoubtedly change the radiological field in the future, Niessen wants to calm concerns that computers will replace humans, saying that human intelligence will complement artificial intelligence in radiology and medical imaging.
When will the future of radiology AI prove its impact in the clinic?
There is a long way to go before we see the real impact of AI in medicine. Referencing a statement that Aliah Sohani made at the RSNA last November, Niessen commented on how the speed of change tends to be overestimated and the impact underestimated. Niessen concedes that, ultimately, the field needs more years of collecting reliable data in order to train the algorithms for AI to improve.
“There's a long way to go, and it takes a while before these new technologies really have an improvement in daily clinical practice,” Niessen says. “In the short term, yes, there is not so much that is really improving the radiologist's daily work, but it will come, and the impact will be very immense.”
He goes on to cite the biggest hurdle to adapting AI for medical imaging is making sure those developing AI are seeking solutions that actually help in the daily clinical workflow of radiologists to improve diagnostics and prognostics. Another large hurdle to the technology’s adoption is ensuring that the machine learning algorithms are able to work in different situations and hospitals, and are able to explain why the algorithms came to a certain decision.
How does Niessen's research contribute to the future of radiology?
Niessen references his own work environment in Rotterdam as a solution to overcoming hurdles in adoption. His team of people working on machine learning and AI are located within the hospitals, directly collaborating with clinical research and radiologists. This collaboration provides access to testing prototypes and receiving feedback to improve the value of imaging for patients.
“The main message that I tried to convey was that AI is an enormous opportunity for the field,” Professor Wiro Niessen reiterated when discussing the talk he had given on the future of radiology AI. “We can learn so much from medical image data, and there’s almost no better way to learn from medical image data as with AI techniques. The promise of precision medicine, to really treat someone in the optimal way, we can try to reach that by incorporating AI technologies, but we’re not there yet. There’s a lot of challenges to overcome, and I also addressed those challenges and how we have to work together with the radiological community and between the people in machine learning in order to try to ensure that AI has a positive effect in medicine.
“AI is disruptive, because it’s really going to automate new things that, so far, we thought only humans can do. So, it’s going to change the field drastically, but I think for a long time to come, human intelligence will complement artificial intelligence, and the same will be true in radiology and medical imaging. I think as long as the focus of the work of the radiologist is to bring the best diagnosis and prognosis to the patient, they can do better with AI. It will be the radiologists with AI that will, at least in the short term, be the future of our field.
“I don’t think we should underestimate the impact of AI in the long term. It will drastically replace the way we are doing diagnosis and prognosis. It will change the way we organize our hospitals, but there is a long way to go, and it takes a while before these new technologies make an improvement in daily clinical practice. So in the short term, yes, there is not so much that is really improving the radiologist’s daily work, but it will come, and it will be immense.
Challenges for the future of radiology AI
“We’ve seen that algorithms are better if you train them with more data. We do not just need the imaging data, we need to describe the data very well. How were they taken? We have to have labels with the data. We have to know what happened to the patient. So these are actually very costly data to obtain, which is why this is a hurdle. But I do think there are also opportunities there. If we start to organize our clinical workflow in a different way, such that we can try to learn from data that are acquired in daily clinical routine, I think we can speed up the process.
“What’s really important for people who develop AI solutions is that they seek solutions that actually help in the daily clinical workflow of the radiologist. That they do help in terms of getting better diagnostics and prognosis, so you really have to prove in the clinical routine that you have value.”
Hear more about machine learning and radiology with Quantib’s automatic brain tissue segmentation solutions. Request a demo today.
