How AI will change MR imaging

Technology & Trends Article 5 Minute Read GE Healthcare Global

 As the late-comer technology to the imaging field1, it’s perhaps ironic that the magnetic resonance (MR) is fostering the next great advancements in diagnostic medicine. The melding of artificial intelligence applications into the soft tissue world of the MRI is achieving remarkable new possibilities.

For example, a team of researchers using 50,000 MRI images recently announced it taught an application to reconstruct better images with less noise than conventional MRI.2 This application is a feed forward platform that reconstructs MRI images immediately to provide near instantaneous diagnosis and treatment protocols. Such expediency is vital in cases of a suspected stroke or cancer – and now possible without invasive diagnostics, such as nuclear medicine.

AI seeks to ID patterns quickly 

According to Dr. Keith Dreyer, Vice Chairman of Radiology Computing and Information Sciences at Massachusetts General Hospital, artificial intelligence (AI) computed imaging applications from MRI and computed tomography (CT) scans have thousands of applications.3 He is working with General Electric Healthcare and Boston Children’s Hospital to develop pattern supporting software. The goal is to develop algorithms to identify commonly reoccurring, life-threatening medical conditions to assist radiologists to yield better and faster diagnostics.

The National Health Service (NHS) in the United Kingdom is deploying a platform to finely tune the thin slices of the MR for prostate tumor4 excise site planning. The AI platform can now accomplish the task in minutes, compared to hours under conventional MRI analysis. It has learned to mark-up organs and the tumors to avoid harming surrounding, unaffected tissue. It results in faster and more precise prostate cancer treatment, according to the radiologists working on the project. The researchers note that the massive amount of data in an MRI scan is a good match for an AI application, which, by nature, is data hungry. 

AI is a good fit for radiologists

Radiologists and other scientists widely agree that MR and associated AI applications are a natural fit with their work. AI and MRI are tools to perform complex diagnostic tasks, while radiologists provide the intelligence to make decisions about the findings.5 The resulting innovation is happening fast in all fields of medicine, most of which require MRI study. For example, New York University is developing software to more accurately predict the onset of type 2 diabetes and associated heart and/or kidney failure and stroke. California researchers have developed an application to detect cardiac arrhythmia – a common MRI diagnostic study – with 97 percent accuracy before the patient suffers a catastrophic event.  

To help physicians keep up with clinical trials deploying imaging AI opportunities, IBM Watson Health has developed an AI tool to find appropriate clinical studies, including MRI studies.6 Another platform enables MRI and ultrasound robotic assisted biopsy and allows more precise and effective MRI/robotic biopsies to diagnose cancer. This technology will couple MRI with ultrasound to create a robot to scan a patient’s entire body in 15-20 minutes, compared to the 45-60-minute scan time in the current standalone MRI protocol. The method is expected to reduce the false-positive rate, which can run as high as 20 percent. Reduced scan time will also lower the cost of the procedure.

Aligning millions of image data points in minutes, rather than hours

Another emerging AI/MRI application is reinventing the process of medical image registration. It is a standard imaging technique in which two images are overlaid to analyze anatomical differences. It is especially useful, for example, in comparing two MRI scans taken several months apart to calculate tumor growth. However, the process requires aligning millions of pixel data points, and a single analysis can take hours.7 Researchers at the Massachusetts Institute of Technology have built an algorithm that reduces the comparison basis to a minute or two.

Dr. Vikram Krishnasetty, a radiologist, and Clinical Director of Information Technology at Columbus Radiology, one the largest radiology providers in central Ohio, said his fellow radiologists need to embrace AI innovations.8 AI will not replace radiologists. But he vigorously makes a case that radiologists who know how to deploy AI will soon be the leaders in the specialty. This should be a natural transition, according to Dr. Luciano Prevedello, Chief of the Imaging Division at Ohio State University Wexner Medical Center. If radiologists could adjust to the advanced technology of MRI and CT over X-ray, he said AI-assisted applications are the next logical step.

AI can extend the life of older MR machines, saving capital dollars

Patients aren’t the only beneficiaries of imaging AI. At the University of Washington, a research team is using AI to give older MRI machines an upgrade to help perform better compared to newer magnetic resonance angiography (MRA). The researchers have applied AI’s deep learning to help MRI mimic MRAs ability to better detail blood vessels associated with aneurysms and other life-threatening conditions. The researchers maintain that such medical pioneering synthesis is a first that will much improve vascular disease diagnosis with older technology already in place.

In another recent study, researchers at Carnegie Mellon University and the University of Pennsylvania are using an AI model to examine MRI scans as a measure to identify patients at risk for suicide.9 The study team trained an AI algorithm to recognize suicidal word ideators based on how the brain lights up during an MRI exam. Not only did the platform pick out suicidal ideators with 90 percent accuracy, but the imaging protocol was able to distinguish between patients who have actually attempted suicide from those who have not. With suicide rates at a 30-year high as of 2014, according to the Centers for Disease Control, tools like this one could have a significant impact.

Buyers Guide_950x250.jpg

Imaging AI will drive better healthcare delivery and policy change

Such AI innovations are not just going to help radiologists realize more time for referring physicians and patient forward duties; they will also be deployed to tackle one of the significant challenges in healthcare – controlling costs through policy change. By 2026, healthcare spending in the United States is projected to leap from $3.3 trillion a year to $5.7 trillion.10 Medicaid, which serves 74 million low-income Americans, cannot tolerate such a jump in funding – it is likely to be politically untenable. AI solutions, such as those emerging in MRI diagnostics, hold great promise in reducing costs through earlier diagnoses, quicker procedure times, and better outcomes.

References:

  1. The History of MR Imaging as Seen Through the Pages of Radiology. RSNA Radiology.  https://pubs.rsna.org/doi/full/10.1148/radiol.14140706. Last accessed July 31, 2018
  2. Artificial Intelligence Enhances MRI scans. HIH Research Matters. https://www.nih.gov/news-events/nih-research-matters/artificial-intelligence-enhances-mri-scans. Last accessed July 31, 2018.
  3. Partners HealthCare and GE Healthcare launch 10-year collaboration to integrate artificial intelligence into every aspect of the patient journey. Partners HealthCare. https://www.partners.org/Newsroom/Press-Releases/Partners-GE-Healthcare-Collaboration.aspx. Last accessed August 9, 2018.
  4. ‘It’s going to create a revolution’: how AI is transforming the NHS. The Guardian.  https://www.theguardian.com/technology/2018/jul/04/its-going-create-revolution-how-ai-transforming-nhs. Last accessed July 31, 2018.
  5. Artificial Intelligence. The Disease Daily.  http://www.healthmap.org/site/diseasedaily/article/future-healthcare-robots-62618. Last accessed July 31, 2018.
  6. Artificial Intelligence. The Disease Daily.  http://www.healthmap.org/site/diseasedaily/article/future-healthcare-robots-62618. Last accessed July 31, 2018.
  7. Faster analysis of medical images. MIT News. http://news.mit.edu/2018/faster-analysis-of-medical-images-0618. Last accessed July 31, 2018.
  8. Artificial Intelligence Is (and Isn’t) Transforming Radiology. COLUMBUSCEO. http://www.columbusceo.com/business/20180416/artificial-intelligence-is-and-isnt-transforming-radiology . Last accessed July 31, 2018.
  9. Controversial Brain Imaging Uses AI To Take Aim AT Suicide Prevention. WIRED. https://www.wired.com/story/fmri-ai-suicide-ideation/. Last accessed July 31, 2018.
  10. Use artificial intelligence to rein in health care costs (Commentary). Syracuse.com https://www.syracuse.com/opinion/index.ssf/2018/06/use_artificial_intelligence_to_rein_in_health_care_costs_commentary.html. Last accessed July 31, 2018.