Radiology and AI - Towards 2030

Professor Eliot Siegel is Professor of Radiology at the University of Maryland School of Medicine in Baltimore, US, and is also an expert in Computer Science. He explained to VISIONS how he sees Artificial Intelligence (AI) integrating with Radiology over the next ten to 20 years.

Speeding up image acquisition

While there has already been a trend to increase efficiency and workflow in Radiology for many years, the COVID-19 pandemic has placed extra emphasis on the need to scan patients rapidly. Artificial Intelligence is currently already used for reconstruction of CT, MRI and PET images.

“We're already using AI to make images higher quality, reduce the time it takes to perform imaging examinations, and to reduce the dose of radiation that is given to patients,” said Prof. Siegel. “The pandemic, in particular, has increased pressure to scan patients more rapidly and AI has been a great solution to allow us to be able to do that.”

The future of improving image acquisition and optimization

AI can be used to generate images. This technology is based on the concept of a generative adversarial network. For example, one algorithm can create fake images, and another one is set to detect the fake, and they can be set to interplay, learn from each other, and improve each other continually.

In a similar way, we can generate higher quality, better images from low-resolution images by using these generative adversarial networks. For MRI, for PET and for CT, we can create images that are higher quality, that require a lower dose, less time to acquire, and allow us to be able to significantly improve the image quality. We can create images that would have otherwise taken hours to reconstruct, in a matter of a fraction of a second.

Artificial Intelligence, or Deep Learning Reconstruction, can already remove the majority of noise to create a high-quality image.

AI trends in different modalities:

MRI: Lower image resolution enhanced to higher resolution image using CNN.
PET: Ten-minute image acquisition reduced to three seconds.
CT: In the next few years, all CT vendors will make major changes in their image acquisition/reconstruction hardware/software to utilize Deep Learning (CNNs).

I think it's much more likely we’ll be using virtual-, and even more likely, augmented-reality.

What will be the Initial “Killer App” for AI Deep Learning in Diagnostic Imaging?

The initial killer apps for AI will be applying deep learning inside the MRI, CT and nuclear medicine scanners.

  • - Major MRI, CT and nuclear vendors will soon adopt Deep Learning to substantially improve image quality, especially texture and reduce scan times and doses.
  • - Iterative reconstruction sacrifices texture for reduced noise, but Deep Learning can optimize image quality without reduction in important diagnostic features.
  • - Model based iterative reconstruction optimizes trade-offs, but is highly computationally intensive, and this has been a major limiting step in its use in day-to-day scanning.

Potential to improve efficiency

One of the most exciting things about AI to Prof. Siegel, who envisages a new paradigm for AI in not only improving productivity, but also reducing stress and fatigue for healthcare professionals, is its potential to improve efficiency.

“Previous research that we have conducted at the Baltimore Medical Center has suggested that radiologists can only spend around 15% of their time actually deciding on a diagnosis and the other 85% mostly calling up images, doing reporting and so on. So, one question is, are there ways to utilize AI to become more innovative and efficient and perhaps much better?” said Prof. Siegel. “Just like most physicians, we're finding that the demand is for faster and faster patient throughput. Particularly with the pandemic, we’re feeling the pressure to be more efficient, and be able to see more patients over a shorter time. And figuring out innovative ways to be able to deal with that is going to be important, and we really think AI can help us in a major way.”

“Within the next ten years, we'll be looking at being able to utilize AI, so that radiologists will be able to increase their reading efficiency from 15% maybe up to 70-80%,” he added.

What do we need from Next Generation AI clinically?

  • Increased accuracy, reliability and confidence, while maintaining efficiency.
  • Affordability.
  • Allows measurement of parameters that couldn’t otherwise be measured, such as liver- or pulmonary texture.
  • Provision of imaging for ‘physical exam’ e.g., bone mineral density on chest CT.
  • Suites of programs.
  • Knowledge of prior studies.
  • Improved efficiency/productivity.

Delivering empathy

An important issue in AI in healthcare is whether it can deliver empathy.
“According to a study done in the Mayo Clinic in 2006, the most important characteristics that patients feel a good doctor should possess are entirely human. According to the study, the ideal physician is confident, empathetic, humane, personal, forthright, respectful and thorough,” remarked Prof. Siegel. “Whether an AI program can be empathetic or not is under exploration.”

Integrated

Prof. Siegel thinks that AI will be increasingly integrated as part of radiologists’ workflow.

“We're going to be using natural language processing to extract information from the electronic medical record, so that we'll be able to have the information we need as we're carrying out image interpretation. And it may be that we’ll move to ‘best-of breed’ AI applications rather than the traditional PACS, or the traditional electronic health record, it may be that we’ll be using more smartphone-like devices that utilize multiple AI applications.
Rather than these applications just doing one off things, such as finding lung nodules or intracranial hemorrhage, we’ll have packages that are used for multiple sclerosis, stroke, trauma etc. that will help us with decision support,” he said. “And I think we're going to be seeing an increasing number of screening studies, as time goes on. I think AI is going to allow us to be able to make predictions and do more intelligent screening studies.”

“One of the things that will be emphasized in the next few years in AI, is not only that it will be used for making findings, but also in looking at quantitative change from prior studies, which is something much more akin to what radiologists actually do,” continued Prof. Siegel.
“I foresee more combination of multiple modalities with AI analysis and ensemble complementary applications working to achieve consensus. In addition, I envisage that patients will begin to have access to radiology analysis packages that are sold as apps but not FDA cleared developed around the world.”

Need for more, not less radiologists

Professor Siegel does not consider AI as a threat to radiology.
“Radiologists do so much more than just make findings. Leaving aside Interventional Radiology (IR), radiologists judge, explain, quality check, counsel, teach, discover, console, explore, create and dozens of other things that computers aren’t even close to being able to do,” he concluded. “We are going to actually need more radiologists rather than less, in the era of AI over the next ten to 20 years. Replacing radiologists with an algorithm is not an option. As Curt Langlotz has suggested, it’s not a matter of AI replacing radiologists, but instead what I think we're going to see is radiologists that embrace the use of AI will replace radiologists that don't.”

Professor Eliot Siegel

Professor Eliot Siegel is Professor and Vice Chair at the University of Maryland School of Medicine, Department of Diagnostic Radiology, as well as Chief of Radiology and Nuclear Medicine for the Veterans Affairs Maryland Healthcare System. He is the Director of the Maryland Imaging Research Technologies Laboratory and has adjunct appointments as Professor of Bioengineering at the University of Maryland College Park, and as Professor of Computer Science at the University of Maryland Baltimore County.

Prof. Siegel graduated in Medicine from the University of Maryland School of Medicine, in 1982. He then completed a Diagnostic Radiology Residency and Nuclear Medicine Fellowship at the same University.

Responsible for the NCI's National Cancer Image Archive, Prof. Siegel served as Workspace Lead of the National Cancer Institute's caBIG In Vivo Imaging Workspace. He has been named as Radiology Researcher and Radiology Educator of the year by his peers as well as one of the Top Ten radiologists.

Under his leadership, the VA Maryland Healthcare System became the first filmless healthcare enterprise in the world. He has written over 200 articles and book chapters about PACS (Picture Archiving and Communication Systems) and digital imaging, and has edited six books on the topic, including Filmless Radiology and Security Issues in the Digital Medical Enterprise. He has delivered more than 1,000 presentations throughout the world on a broad range of topics involving computer applications in imaging and medicine. Prof. Siegel served as symposium chairman for the Society of Photo-optical and Industrial Engineers (SPIE) Medical Imaging Meeting for three years and is currently serving on the board of directors of the Society of Computer Applications in Radiology. He is a fellow of the American College of Radiology and of the Society of Imaging Informatics in Medicine.

“It’s not a matter of AI replacing radiologists, but instead what I think we're going to see is radiologists that embrace the use of AI will replace radiologists that don't.” – Prof. Eliot Siegel

“One of the most exciting things about AI is its potential to improve efficiency and productivity.” – Prof. Eliot Siegel


Disclaimer: The opinions expressed in this material are solely those of the SME and not necessarily those of Canon Medical Systems.
Canon Medical Systems does not guarantee the accuracy or reliability of the information provided herein.