New Rule for Medical Imaging: AI Inside

Before it became fashionable to digitize medical records, medical imaging had already moved to digital with a passion. Long the gold standard for non-invasive detection and analysis of the inner workings of the human body, medical imaging technology has relentlessly advanced by boosting resolution, expanding data dimensionality from pixels to voxels to streams of moving images, and miniaturizing systems to greatly expand access in remote and lower cost settings. The resultant digital imaging wave was likely every healthcare system’s first big data wakeup call as basements full of films and plates were replaced with data ‘lakes’ full of disks and image files.

While the sophistication and quality of the digital image ramped, the population of radiologists and the average amount of time they have to decide each case has been largely unchanged. Advancing computer, storage and networking speed inside the imager has allowed the imaging suite to pretty much keep pace, but the clinician has been squeezed hard. They are under greater and greater pressure to maintain diagnostic quality without slowing throughput. With daily caseloads and the number of images per case increasing, errors, especially in interpretation of images which is so critical to accurate and timely diagnosis, will likely increase, putting patients at more risk.

This is where artificial intelligence (AI) solutions can be a game-changer, as its rapid adoption by medical imaging’s global leaders demonstrates. AI models use neural networks that can mimic many of the human processes used to classify, recognize, and analyze new and highly complex sets of inputs. When used with and compared to vast quantities of already captured & annotated historical images, these AI models can be trained to recognize the same anomalies as fast or faster than a human can. As a result, computing systems with AI inside can go beyond just showing the image to the clinician; these systems can now prep the image with relevant markers and metadata critical for an effective and now more efficient, diagnostic review by the human expert.

AI-embedded X-ray system is designed to identify pneumothorax with high accuracy.

In addition to enhancing and speeding the diagnostic review itself, AI innovation can also positively impact workflow and resource utilization, allowing clinicians to keep pace with the ever growing volume of cases and the complexity of image data sets they contain. For example, based on what the models identify in the images, these solutions can prioritize the most time critical cases alerting the radiologist and placing those related images at the front of the queue. The result? Higher productivity, less burnout, reduced time to critical interventions and, I believe, improved quality of care.

Deep Learning-based models demonstrate bone age, and lung segmentation at speed needed for critical case triage.

Other opportunities for AI can be found beyond localized use. Combining radiology datasets across institutions can enhance AI models where more heterogeneity of datasets creates more robust, less biased models. AI assisted radiology can help manage remote care delivery priorities by screening out normal cases, and using telemedicine to connect abnormal cases to expert clinicians either remotely or in clinic. AI tools can also harness and make sense out of the vast amounts of research now being generated at a rate much greater than human experts are capable of, and build this upon knowledge systems that can update quickly.

AI-based cardiac MRI inferencing demonstrated ability to deliver results to technologists cardiologists and radiologists in real-time.

This is not some future vision or aspirational scenario; the technology and systems required to enable these capabilities are available and in use today. By providing the computing technology, software and AI tools such as the OpenVINO™ toolkit, Intel has been able to help usher in the growing use of AI to positively impact such a wide array of imaging modalities and applications, even challenging the status quo by working to enable state of the art inferencing on mainstream CPUs without the need for expensive, specialized accelerators.

At the dawn of the internet, legendary Intel CEO Andy Grove declared that “all business will be e-Businesses.” Care delivery is no exception and stands to be one of the most dramatically transformed by AI-enhanced information that will amplify, focus and extend the critical expertise of specialists, like radiologists. Using AI-enhanced approaches to traditional imaging diagnostic models and processes is just one example of how this powerful new technology realm will drive faster time to insight, greater workflow and process efficiency, and better access to specialized expertise without sacrificing patient safety and outcomes. Healthcare’s digital transformation has just begun, and AI promises to quicken the pace of change for the better – in imaging and much, much more.

Learn more about how Intel and the vast Intel-based ecosystem are working together to transform healthcare with data-driven innovation.

*Other names and brands may be claimed as the property of others.

 

Published on Categories Artificial intelligence, Internet of ThingsTags , , , , , , ,
David P. Ryan

About David P. Ryan

Dave leads the global Health & Life Sciences business unit at Intel that focuses on digital transformation from edge-to-cloud in order to make precision, value-based care a reality. His customers are the manufacturers who build life sciences instruments, medical equipment, clinical systems, compute appliances and devices used by research centers, hospitals, clinics, residential care settings and the home. Dave has served on the boards of Consumer Technology Association Health & Fitness Division, HIMSS’ Personal Connected Health Alliance, the Global Coalition on Aging and the Alliance for Connected Care.

Leave a Reply

Your email address will not be published. Required fields are marked *

Name *

This site uses Akismet to reduce spam. Learn how your comment data is processed.