Real-World Applications of AI

Professor Stefan M. Niehues, MD, MHBA, Professor Mickaël Ohana, MD, PhD, Dr. Benoît Sauer, MD

Specialists across the world have welcomed Canon Medical’s growing portfolio of Deep Learning Reconstruction (DLR) technologies into their daily practice. Three leading radiologists share their experiences on the performance of real-world applications already available via Canon Medical’s Altivity.

Radiology is a key area for AI innovation. AI applications that are already available and in clinical use include image optimisation through e.g. smart workflow automation, image reconstruction, automatic stroke detection, the detection of pulmonary embolism, aortic dissection in emergency situations, and more.

Canon Medical is at the forefront of developing AI-based healthcare solutions that deliver quicker and deeper clinical insights and support clinical decision-making. Launched in 2021, Altivity, Canon Medical’s suite of cutting-edge AI technologies, supports more informed healthcare workflows and fast and tailored patient care. Our advanced AI-based systems have been developed together with some of the world’s leading experts.

Professor Stefan Niehues, Radiologist, Deputy Director, and Senior Physician at the Campus Benjamin Franklin of Charité University Hospital, in Berlin, Germany, explained how Canon’s AI applications perform in emergency CT settings

“AI has merged into our routine clinical workflow. The typical application areas include stroke, pulmonary embolism, and aortic dissection - everything which is important to recognise very fast and to react on those findings,” he remarked. “We started using Canon’s Automation Platform for stroke and have extended it. It provides immediate assistance. You can use it to take a ‘second look’ or to improve confidence.”

“In the conventional process of compiling a report, you start with the image acquisition and transfer it to some kind of IT technology. In post-processing, you may need to select or load further images, analyse them, save example images or screenshots, prepare them for a result, and then have the final result written and transferred to your other disciplines. It's quite a long process.” he explained. “The Automation Platform really shortens this because the categorisation, and prioritisation of acquired images for post-processing are all performed automatically with a zero-click solution. After the scan and image transfer to the AI, you will get the final results for further interpretation without further ado, which will then reduce the burden of work, increase productivity and save time. Not only your time, but also the time for your patients until they get the final results.”

The applications for the Automation Platform have broadened since its introduction.

“It started with a stroke solution with hemorrhage detection, the ASPECTS score, a fully automatic perfusion calculation and visualisation, and large vessel occlusion detection, and now applications to detect pulmonary embolism (PE) and aortic dissection (AD) have been released,” said Professor Niehues.

PE in particular is a common but sometimes fatal disease that can present with a wide range of symptoms from none at all up to sudden death. Many patients can have mild or no symptoms, but still need to be diagnosed and treated.

Aortic dissection is not common, and mostly presents as an acute condition in patients with catastrophic illness. However, early and accurate diagnosis and treatment is crucial for patient survival. CT is the most common modality of choice because of its widespread availability in emergency departments.

“The Automation Platform flags no findings of aortic dissection, or in the case of positive findings, it will indicate the presence and its location of the dissection. In the case of multiple dissections, it provides multiple key images so you can see the whole extent,” said Prof. Niehues. “Again, here, the median processing time is a little over 34 seconds, so it's very, very fast. And there is an even greater sensitivity of 96% and a specificity of 97%. You can really rely on the results being presented by the Automation Platform.”

Patients with coronary artery disease can benefit from use of Canon’s Super-Resolution Deep Learning Reconstruction (SL-DLR) technique, PIQE. Professor Mickaël Ohana, Consultant Radiologist at the Strasbourg University Hospital in France, explained how it provides clinicians with the possibility for higher diagnostic confidence and clarity as compared to conventional image reconstruction approaches in visualising small arteries, plaques, and fine cardiac structures, and has the potential to assist clinicians in coronary atherosclerosis patients' cardiovascular risk stratification.

“PIQE directly brings the potential advantages of Ultra High Resolution CT (UHR-CT) to conventional CT. It is about merging the best of both worlds,” said Professor Ohana.

“Current research is focused on the advantages induced by very thin slice-thickness and increased matrix size. Mostly a sharper image quality, reduced artifacts, partial volume and blooming, and also an improved detection and characterisation of small anatomical structures,” he continued. “Through this, Super Resolution CT has shown promise in correctly identifying non-obstructive diseases that were labeled as obstructive with conventional CT.”

PIQE is currently available for cardiac CT.

“We have noticed significant noise reduction, and increased sharpness of all the vascular and anatomical structures with PIQE compared to deep learning or hybrid IR reconstructions. The conspicuity of the structures and the contour can be more easily seen. Even in lesions with very subtle arterial calcification, PIQE provides a better view. This is the same with curved MPR - the sharpness of the vessels is better with PIQE. You can also achieve better delineation of smaller arterial branches. And you get an increased conspicuity of calcifications, mostly on faint or subtle low-density calcifications,” remarked Prof. Ohana. “All these points - the noise reduction, the increased sharpness, better delineation of small structures, and the increased conspicuity of calcifications - lead to a higher image quality with PIQE”

“In the end, in routine clinical work, you can expect to get the advantages of the Super Resolution CT images, but without the drawbacks,” he added. “And at a lower cost with, of course, a higher number of machines to be able to do that. The availability of a wide area detector, which is something that once you get it, you cannot go back, when you do cardiac imaging. The faster rotation time, the ability to use systematically for all patients on 100kV and all that at the lowest radiation dose. This is really like combining best of both worlds.”

“For now, this technology is dedicated to cardiac CT, but it will for sure come to chest CT, MSK, and Neuro later. What is important is that you can use it without any impact on the workflow or the dosimetry. You have an increased perceived image quality of coronary CTA, and in the end, it could carry a potential diagnostic impact as we have seen, probably a better assessment of highly calcified vessels, possibly also a better delineation of minimal atherosclerotic lesion, and also maybe a possible better plaque quantification,” he concluded. “It is very promising, and I think it's only the beginning of the technology. We hope that we will see more in the future.”

“Deep Learning Reconstruction is a major additional layer for improvement of image quality and also time in MRI, which is very important for patients,” said Dr. Sauer. “In addition, radiologists have more choice and more capacity to adapt to make better images for each patient characteristic.”

In abdominal imaging, quality is essential to make an accurate diagnosis. For upper abdomen scanning, the length of the patient’s breath-hold is a limit. For whole body diffusion scanning, the duration of the examination can be problematic for some patients.
For whole-body DWI diffusion, for myeloma, comfort is very important. For musculoskeletal imaging, especially trauma- related scanning, total scanning time needs to be fast.

“With Canon’s Deep Learning Reconstruction techniques, AiCE and PIQE, we now have an MRI liver protocol of less than ten minutes, complete with a low dose CT scan of the whole thorax-abdomen. For unwell patients experiencing difficulties with breath-hold, we make shorter sequences of 13 seconds in dynamics, so we have high quality. For prostate, we can perform an examination in ten minutes,” said Dr. Sauer. “It works also for other things.

For example, musculoskeletal imaging and knee trauma with a rupture of LCA, and bone marrow oedema. The total examination is less than six minutes. The advantage is quicker results with fewer movement artifacts, so it's very important. We can obtain good quality images as well as a good diagnosis, fast.”

Canon Medical is developing intelligent technologies to make a whole new level of quality, insight, and value across the entire care pathway possible. With these insights, it is becoming apparent that AI-based healthcare support can be successfully embedded into clinical practice for the benefit of patients and clinicians. //

---