Expert Perspectives on AI

Professor Mathias Prokop, MD, PhD and Professor Mickaël Ohana, MD, PhD
Two of the world’s leading radiologists give their insights into Artificial Intelligence (AI) in medical imaging, including their experiences of the performance of new applications currently available via Canon Medical’s Altivity platform. Altivity is our approach to AI innovation that uses intelligent technologies to create a whole new level of quality, insight and value across the entire care pathway.
Professor Mathias Prokop is a Radiologist and the Head of the Department of Medical Imaging at Radboud University Medical Center in Nijmegen and University Medical Center Groningen, the Netherlands. As a leading figure in research, he often shares his perspectives on how the latest systems perform in practice.

We asked him to give his insight into how AI is currently used in real-world scenarios in radiology departments.

“At present, AI is used in many radiology departments in a way that many people don't realize that they're actually using it,” he remarked.
“The most successful techniques currently in everyday use include reconstruction in CT, which utilizes deep learning technology to improve image quality by reducing noise and shorten reconstruction time. Deep Learning Reconstruction algorithms, such as Canon’s Advanced intelligent IQ-Engine (AiCE), are used in many of our scanners, but most radiologists don't realize it's actually AI that is in the background.”

“There are also AI systems that are focused on interpreting images,” he continued.

“AI technologies that are successful require the least interaction. At the moment, a lot of radiology departments are trying out AI and are attempting to figure out which applications work well.”

Professor Mathias Prokop, MD, PhD

Streamlined workflow

Professor Prokop believes that AI will ultimately help us become more effective and reduce those tasks that we are either not good at, like quantification, or tasks that we don't really like, such as repetitive ones - for example, looking for pulmonary modules in lung screening or many images in mammography screening, or in oncology scanning, in which follow-up is required.

“Ensuring that tasks which are still time consuming and relatively repetitive are condensed and completed faster means that we can use the time gained for things that add value, such as consulting functions, spending more time on more difficult cases, and giving suggestions about the next steps in the diagnostic process, or for the right therapeutic procedures,” he said.
Professor Prokop believes that the best route to commercial success for AI systems is through streamlining workflow.
“Scalability is an issue in financing AI technology in the healthcare setting,” he said. “For widespread uptake, AI systems need to massively improve outcomes or enable the radiologist to complete their work in a shorter period of time. I think those that mainly focus on workflow will become commercially successful first.”

This 52-year-old man with a history of smoking, dyslipidemia and a family history of CAD presented with typical angina and a positive stress test.
A coronary CTA was requested to rule out coronary artery disease and determine the optimal treatment based on the results of the ISCHEMIA trial (https://www.nejm.org/doi/full/10.1056/nejmoa1915922). The calcium score scan showed mild coronary artery disease (CAD) with an Agatston score of 34. A mixed plaque seen in the proximal LAD. PIQE reveals better delineation of the plaque and less blooming of the calcification compared to AIDR* 3D

*: Adaptive Iterative Dose Reduction

The role of Altivity

Professor Mickäel Ohana, Consultant Radiologist, at Strasbourg University Hospital, France, has also extensively researched the use of AI in clinical practice. He has collaborated with Canon Medical for many years and explained how Altivity applications are now integrated into everyday workflows.

“Altivity is now routine for us,” he remarked. “Five years ago, we spoke of AI as something that might replace radiologists, rate the images, make analyses and predictions, but what we've actually seen is that through Deep Learning Reconstruction, Altivity applications can work on all images, at any time, to improve the image quality, to lower the radiation dose, and to ease our diagnosis. So, this is something that is routinely available for every examination and really makes a difference.”

Professor Ohana also has experience with Automation Platform – Canon’s AI-based, zero-click solution that uses deep learning technology, aiming at automating clinical workflows.

“We have found Automation Platform integrates well in PACS and works quite seamlessly in the background,” he said. “The aim here is not to have something complicated, but to be something really integrated into your everyday workflow quite nicely in the background, so that it won't take your time, but it will maybe create more time for you.”

A leap forward

What stands out with advances in Canon Medical’s Altivity approach is that there is a real leap forward in comparison with other reconstruction techniques. Prof. Ohana has evaluated a new Super-Resolution Deep Learning Reconstruction (SR-DLR) named Precise IQ Engine (PIQE) for Cardiovascular CT examinations.

“The difference is in sharper images with increased spatial resolution, and with less image noise. Combining both is really a tremendous step forward and this is something that increases the image quality a lot,” he said. “This is what really struck me most in the beginning.”

The Strasbourg University Hospital installed an Aquilion ONE / PRISM Edition from Canon more than two years ago. The system provides an ideal platform to continue to develop and explore AI capabilities for Professor Ohana and his team.

“I think the strength of the system is that it can do everything well. It is performing extremely well thanks to AiCE DLR, which has been introduced subsequently,” he said. “That provides a real step forward not only in increasing image quality and spatial resolution, but also decreasing the dose. So, altogether, it makes a great all-round CT and for carries out everything quickly and well.”
“With the Aquilion ONE / PRISM Edition, we have the ability to perform Spectral Imaging with DLR,” he added. “This is a nice addition to an already capable CT. I use this mostly for arterial enhancements in CTA, and I believe that in some cases it really helps us – mostly, for example, if the patient has a limited readout function and we want to lower the dose of contrast that we give. I also have colleagues that use Spectral Imaging for visceral purposes in which you can increase your sensitivity. This is also a nice addition to the workflow.”

New possibilities

The Strasbourg University Hospital installed an Aquilion ONE / PRISM Edition from Canon more than two years ago. The system provides an ideal platform for Professor Ohana and his team to explore the capabilities of AI.

“I think the strength of the system is that it can do everything well. It is performing extremely well thanks to AiCE DLR, which has been introduced subsequently,” he said. “That provides a real step forward not only in increasing image quality and spatial resolution, but also decreasing the dose. So, altogether, it makes a great all-round CT for everything and carrying out everything quickly and well.”

Now, with advances in Deep Learning technology, come new opportunities to explore more. Additionally, with improvements in image quality afforded by AiCE and PIQE, there are new possibilities in Spectral Imaging.

“With the Aquilion ONE / PRISM Edition, we have the ability to perform Spectral Imaging with DLR,” he added. “This is a nice addition to an already capable CT. I use this mostly for arterial enhancements in CTA, and I believe that in some cases it really helps us – mostly, for example, if the patient has a limited kidney function and we want to lower the dose of contrast that we give. I also have colleagues that use Spectral Imaging for visceral purposes in which you can increase your sensitivity. This is also a nice addition to the workflow.”

Future expectations

Now that they have gathered experience with the AI-based systems of today, both experts have been able to formulate clearer views of the future.

“When you look at, say a period of five to ten years ahead, I believe we will see systems become increasingly autonomous, so they will pre-read images and pre-populate our reports in a way that we can just check them. This will mean that we will be able to do things much faster,” said Professor Prokop. “What we would like to see in the intermediate future, are AI systems that help us improve our workflow much further and become even more effective in reducing boring tasks towards ensuring that we can use our time to do things that are more interesting and add more value for our clinical colleagues and patients.”

“Our hope is that AI will help us by giving us more time for more advanced and complex cases, so that we can focus on what is really important and what has higher value”

Professor Mickäel Ohana, MD, PhD