E-Bulletin, Issue 1 from March 31, 2020

AI radiology market to reach US $3.5 billion by 2027; More radiology departments implementing AI this year

Welcome to our introductory monthly newsletter on neuroradiology and AI. We hope you find it helpful. Please feel free to share it with others.

If you would like to be informed when the new issue is out, please sign up here.

Study: Deep learning in neuroradiology: A systematic review of current algorithms and approaches for the new wave of imaging technology

This review in Radiology: Artificial Intelligence delves into 155 published studies to identify and describe the technical characteristics of deep learning applications for CT, MRI, PET and ultrasound imaging. It argues that while the field of artificial intelligence in neuroradiology has grown quickly in the past several decades, the research needs more reproducibility, research standardization and generalizability before fully applying it to clinical practice. Authors analyzed peer-reviewed studies published in English, spanning January 1, 2000 through September 1, 2019.

The largest imaging category in the 155 studies was MRI, with 115 studies (74% of the total), focusing on algorithms for image classification, segmentation, generation, detection, reconstruction and prediction. The algorithm usage varied widely, and included determination of brain age and prediction of Alzheimer’s disease, and ischemic stroke lesion segmentation.

The other imaging modalities had fewer studies. Functional MRI had 19, on evaluations for brain activity during rest, mild cognitive impairment, and sensorimotor tasks. Nine CT studies were referenced, all two-dimensional or three-dimensional convolutional neural networks (CNNs), for tasks like classification, detection, image generation, segmentation, and prediction. The purposes were reconstructing CT data, Alzheimer’s disease, traumatic events and identifying brain changes after hemorrhage. The review noted 18 PET studies, with many including multimodal analysis of PET and MRI, for classification, generation, detection, prediction, and reconstruction, mostly for Alzheimer disease and amyloid burden classification.

Most applications were for research or architectural development, and a small number had begun clinical trials. The authors concluded that there was no standardized reporting scheme for the machine learning models, making interpretation, direct comparison and meta-analysis challenging. They also called for researchers to open the models and code for their algorithms so other researchers could reproduce, evaluate and test external data for validation.

Study: Analysis of head CT scans flagged by deep learning software for acute intracranial hemorrhage

This study in Neuroradiology analyzed an academic medical center’s implementation of deep learning software to detect acute intracranial hemorrhage on non-contrast head CT (NCCT), and to prioritize the worklist, in various clinical settings. Neural network software from Aidoc in Tel Aviv was used to review urgent NCCT scans. They included all scans flagged as positive for acute intracranial hemorrhage on the neuroradiology worklist, to prospectively include in assessment. They classified scans for presence and type of hemorrhage, initial or follow-up scan, and patient visit location (trauma/emergency, inpatient, and outpatient departments). Authors concluded that the AIDOC software detection varied based on patient visit location. Many of the cases were follow-up exams, including in-patient exams, which can help them optimize the AI-drive clinical workflow.


Brain Pathology Assessment with Panoramic Perfusion

Neuroradiology and AI in the news

Artificial intelligence radiology market expected to reach US $3.5 billion by 2027: Research by Absolute Market Insights estimates that the AI radiology market will reach US $3.5 billion by 2027, with an estimated CAGR growth of 16.5% over the forecasted period, as CT adoption rises. They estimate the 2018 global AI radiology market was US $187.61 million. “Increase in usage of artificial intelligence for neuroradiology to diagnose abnormalities in head, brain, spine and neck is driving the market.”

Imaging report shares AI adoption rates and best practices: A new KLAS report, Artificial Intelligence in Imaging, shared best practices for AI adoption from industry users. Currently 46% of organizations say they are currently adopting AI, 38% are making plans, and 16% are not making plans to implement AI. In organizations already adopting AI, 48% plan to have a high depth of adoption, with multiple use cases and more imaging departments than before. Of those adopting, 43% plan for a medium level of adoption, while 8% plan for a low level. The report also shares interviews with 81 technology companies, asking which are best positioned to adopt AI in 2020.

The American Board of Artificial Intelligence in Medicine (ABAIM) incorporates: The ABAIM incorporated as a not-for-profit entity, and plans to credential healthcare professionals and others seeking to understand the role of AI, machine learning and deep learning in health care.

Enhancing neuroimaging with artificial intelligence: Neuroradiologist Suzie Bash MD explains in Applied Radiology how AI is currently used in neuroimaging. “Quantitative volumetric MRI tools add diagnostic value, accuracy and efficiency. They assist in eliminating reader subjectivity and provide objective longitudinal assessment of disease…AI applications are quickly becoming embedded in the fabric of advanced, state-of-the-art neuroimaging by creating value, enhancing quality and saving time.”

RadNet to acquire DeepHealth: RadNet, a U.S. outpatient diagnostic imaging firm with 290 centers, is acquiring DeepHealth for its AI-based imaging analysis technology. Their technology currently focuses on 3D mammography. DeepHealth was co-founded by neuroradiologist Gregory Sorensen, MD. “We are more certain today than ever before that artificial intelligence will transform the diagnostic imaging and radiology industry. Machine learning, big data applications and automation algorithms will allow us to deliver our services more cost effectively, efficiently and accurately,” said Howard Berger, MD, RadNet Chairman and Chief Executive Officer in a statement. RadNet also owns New Logic, which develops AI tools for radiologists, and is partnering with Whiterabbit Artificial Intelligence for operational programs.

With more and more AI algorithms being validated, AI becomes a necessary part of an efficient radiology assessment and workflow. Today’s world has shown how important it is to have the tools at hand that can help healthcare professionals and institutions to deliver the best possible patient care when each minute counts.

Would you like to learn more about how AI works and how it can be implemented in your department, feel free to reach out to our experts!