Exploring the Potential of AI for Clinical Practice

Radboud University Medical Centre (Radboudumc), in Nijmegen, the Netherlands, has been at the forefront of medical imaging for decades. With a strong team of experts, its Department of Medical Imaging leads the world in many elements of research across all imaging modalities. Over the years, it has worked together with Canon Medical in evaluating its new technologies and techniques. Most recently, it’s Diagnostic Image Analysis Group (DIAG) has been involved in the development and implementation of advanced imaging techniques that feature AI-solutions, in collaboration with Canon. Dr. Anton Meijer, Neuroradiologist, and Dr. Ruud Becks, Radiologist, explain what their evaluation of Canon’s Automation Platform has already shown.
VISIONS spoke with Dr. Anton Meijer, Neuroradiologist, and Dr. Ruud Becks, Radiologist, at the Radboudumc in Nijmegen, the Netherlands.

Focus on stroke

Alongside clinical work, the neuroradiologists at Radboudumc collaborate on research projects to study various neurovascular and neurodegenerative diseases with clinical partners in Neurology and Neurosurgery Departments. One key focus is on the use of advanced neuroimaging tools to improve detection, diagnosis and treatment of stroke.

In acute stroke, imaging needs to be performed and interpreted immediately, so that the right treatment for each patient can be determined and implemented fast.

“In a stroke center such as ours, a 24/7 service is needed for a fast and accurate diagnostic work-up of patients presenting with an acute neurological deficit. This is necessary to identify treatment pathways and improve the clinical outcome for individual patients,” explained Dr. Becks. “However, this puts a great demand on our neuroradiology service, with the need for advanced imaging techniques, such as brain CT perfusion and 4D-CTA.”
VISIONS spoke with Dr. Anton Meijer, Neuroradiologist, and Dr. Ruud Becks, Radiologist, at the Radboudumc in Nijmegen, the Netherlands.

Potential for powerful processing

The greatest demands in acute stroke imaging workflow concern the automated processing of brain CT perfusion, and in supporting the detection of intracranial hemorrhage and large vessel occlusions. Timely and adequate treatment in stroke is crucial to achieve vessel revascularization as soon as possible and to minimalize the extent of brain tissue damage.

Radboudumc carries out specialist neuroimaging techniques which are made possible through the Centre’s Canon Medical CT systems, including an Aquilion ONE / PRISM Edition wide-detector CT and an Aquilion Precision high-resolution CT scanner.
Left: Dr. Ruud Becks, Radiologist, Radboudumc. Right: Dr. Anton Meijer, Neuroradiologist, Radboudumc.
“Our Canon CT systems enable us to perform state-of-the art, dynamic, and high resolution imaging of neurovascular anatomy and pathology,” said Dr. Meijer. “And with a dedicated Vitrea Advanced Visualization workstation for data-processing, we can read these advanced imaging studies easily. Deep learning based image reconstruction (Advanced intelligent Clear-IQ Engine (AiCE)) is routinely used in our clinical practice, and contributes to superior imaging quality with lower noise and improved tissue differentiation. This also enables a reduction in radiation dose.” “Automated processing of brain CT perfusion can offer time savings and provide crucial information in order to confirm the diagnosis of ischemic stroke and to estimate tissue viability,” added Dr. Becks. “Support in the detection of intracranial hemorrhage and large vessel occlusions is of added value in treatment decision-making in the acute setting, especially with regards to intravenous thrombolysis and mechanical thrombectomy.”

“Abnormalities can be subtle but are clinically relevant, e.g. a small amount of hemorrhage, or an occlusion of a smaller middle cerebral artery branch. There is a great need for workflow support and for making advanced imaging studies easy to read,” remarked Dr. Meijer. “This could be achieved by the implementation of an automated workflow and integration of AI-tools, such as the Automation Platform.”
Figure 1: Non-contrast brain CT demonstrating a small amount of subarachnoidal hemorrhage in the central sulcus on the right side.
Figure 1: Non-contrast brain CT demonstrating a small amount of subarachnoidal hemorrhage in the central sulcus on the right side.
Figure 2: Head CTA demonstrating a short segment occlusion of a middle cerebral artery branch on the left side. The Automation Platform could aid in the detection of such a vessel occlusion, which is of clinical relevance as this would be target for mechanical thrombectomy.
Figure 2: Head CTA demonstrating a short segment occlusion of a middle cerebral artery branch on the left side. The Automation Platform could aid in the detection of such a vessel occlusion, which is of clinical relevance as this would be target for mechanical thrombectomy.

Promising results

The Medical Imaging Department is currently evaluating Canon’s Automation Platform (AP) for CT and has found it so promising that it is planning to implement it into its clinical workflow.

“We anticipate a faster and improved workflow in acute stroke imaging with the AP, as well as support for the radiologist in making the correct diagnosis and to aid the clinicians in treatment decision-making.” said Dr. Meijer.
Data sharing and cross-hospital workflow are of equal importance in order to facilitate treatment decision-making.

“Quick access to the imaging studies is important for the neurologists and neuro-interventionalist for the decision to treat the patient with intravenous thrombolysis, or to transfer the patient to the angio-suite for mechanical thrombectomy,” explained Dr. Meijer. “Efficient communication with exchange of imaging studies between hospitals is of relevance for the decision whether or not to transfer a patient to a specialized stroke center.” “Canon’s AI-tools provide key information for worklist prioritization and to support the radiologist in reading the CT studies. We particularly like the AP’s Insight Results which provide a quick and clear overview of the imaging findings of NCCT, CTA and CT perfusion,” added Dr. Becks.
Figure 3: Brain CT perfusion of the patient in Figure 2 demonstrating a perfusion deficit in the left hemisphere with a pattern of infarct core with surrounding penumbra.
“A unique perfusion algorithm is now also available with AP - Bayesian perfusion. This is a probabilistic estimation of the perfusion values and is also delay insensitive. This computation method is less sensitive to noise and enhance the diagnostic liability of ischemic stroke detection. In addition, the Bayesian method can be combined with the AI based image reconstruction algorithm “AiCE” from the acquisition side to reduce the noise even more,” said Dr. Meijer. “The Bayesian algorithm provides superior quality CT perfusion maps, with excellent differentiation between grey- and white matter structures. It has been demonstrated that it can be reliably used for patient selection based on tissue viability estimates, which is of relevance in the extended time windows (up to 24 hours after onset of symptoms).”
Figure 3: Brain CT perfusion of the patient in Figure 2 demonstrating a perfusion deficit in the left hemisphere with a pattern of infarct core with surrounding penumbra.

“We anticipate a faster and improved workflow in acute stroke imaging with the Automation Platform, as well as support for the radiologist in making the correct diagnosis and to aid the clinicians in treatment decision-making.”

Dr. Anton Meijer, Neuroradiologist, Radboudumc.

Dr. Anton Meijer is a radiologist at the Department of Radiology of Radboud University Nijmegen Medical Centre, Nijmegen, the Netherlands, with specialization in neuro-, head and neck, and emergency radiology. He participates in clinical and research projects in imaging in neurovascular and neurodegenerative diseases

Easy-to-use

The added value of ease-of-use provided by the AP is potentially highest for less experienced colleagues, such as general radiologists and radiologists in training.

“Automated detection of pathology on imaging studies can also support worklist prioritization, where a study is highlighted when an abnormality is detected. These studies can be read with priority by the radiologist,” remarked Dr. Meijer. “The initial test feedback from the radiologists who are experienced in stroke imaging is that the AP is really easy and intuitive to use and requires little training. These aspects are important and will be tested further as we roll out the solution to other staff who will use the solution in earnest in a clinical setting,” said Dr. Meijer. “It is crucial that the AP is implemented in clinical practice, so we can share our ‘real-world’ experiences.”

Managing expectations

It is important that users are well instructed about the intended use of an AI-tool in order to manage expectations.

“For example, the LVO-detection tool is trained for the detection of large vessel occlusions in the anterior circulation on conventional CT angiography. This means that distal vessel occlusions, and vessel occlusions in the posterior circulation (e.g. basilar artery thrombosis) cannot be detected by the AI-tool. Therefore, the AI-tool does not replace the CT angiography evaluation by a radiologist, and does not replace the radiologist supervising a resident. The results of the AI-tool need to be verified by the radiologist, and should be overruled in case of a false-negative or false-positive result,” said Dr. Meijer. “The end-users should be familiar with the fact that AI-models can fail to generalize when applied in clinical practice with heterogeneous populations and imaging protocols, which can result in a lower than expected diagnostic performance. Furthermore, interaction with existing systems and human-AI interactions may provide different results than reported performance based on prior validation studies. It is therefore advised that the usage and the performance of the Automation Platform is monitored, to ensure that it meets the demands of the users and to further improve its performance and to incorporate new imaging features (e.g. ASPECTS and collateral scores).”

“ Automated processing of brain CT perfusion can offer time savings and provide crucial information in order to confirm the diagnosis of ischemic stroke and to estimate tissue viability.”

Dr. Ruud Becks, Radiologist, Radboudumc

Dr. Ruud Becks is a Radiologist at the Department of Medical Imaging of the Radboud University Nijmegen Medical Centre, Nijmegen, the Netherlands. He is currently completing a fellowship in neuroradiology with special interest in neurovascular imaging and protocol optimization. In cooperation with Dr. Anton Meijer, Dr. Becks is involved in the evaluation of different AI-solutions in Stroke.

Great opportunities

Over the last decade, the Medical Imaging Department at Radboudumc has developed a close collaboration with Canon Medical Systems in the development and clinical implementation of new hardware and software imaging solutions, mainly in the field of CT. These are applied to different organ systems, including the abdomen, thorax, musculoskeletal and brain, in collaboration with clinical and preclinical experts in the field.

“Our primary focus has been evaluating the Stroke CT Package in the AP, but there are great opportunities to integrate a similar workflow for other organ domains, such as cardiovascular imaging, traumatology and oncological follow-up,” said Dr. Meijer. “I think one thing to recognize is that AI is going to change practice and I think in neuroradiology, acute stroke is one of the main subjects in which AI will definitely contribute. In the end, I think it will really improve the speed and the quality of the diagnosis in acute stroke, but will not replace the radiologist.”
The new entrance of Radboudumc.

sou

Automation Platform, Powered by Altivity
Canon Medical’s Automation Platform (AP) is an AI-based, zero-click solution that uses deep learning technology to streamline workflow for fast, actionable results every time. From scanner to clinical decision, support is provided by leading-edge deep learning technologies that process and deliver images for accurate triage, worklist prioritization and treatment decisions. In parallel with the AP, Canon has developed the Stroke CT Package - a suite of AI-based applications designed to streamline strokerelated workflow. By automatically consolidating results into a single summary and alerting for abnormalities, the AP helps to support fast triage to facilitate treatment decisions.

The Stroke CT Package helps the specialist to swiftly analyze and evaluate images, and detect the signs of ischemic and/or hemorrhagic stroke in minutes. This ‘one-stop’ solution provides access to information required to administer life-saving treatment for patients.