Mass General Brigham and the future of AI in radiology
Credit: David Fox
Artificial intelligence is making fast progress in the field of radiology. Clinical adoption of AI by radiologists has gone from none to 30% from 2015 to 2020, according to a study by the American College of Radiology.
At the high-profile health system Mass General Brigham, clinicians and IT professionals are working together to advance the use of AI and machine learning in radiology. They're making great strides in making the practice of radiology better for radiologists and health outcomes better for patients.
Dr. Keith Dreyer is chief data science officer and vice chairman of radiology at the Mass General Brigham health system. He also is associate professor of radiology at Harvard Medical School and a member of the American College of Radiology Board of Directors.
Healthcare IT News interviewed Dreyer to learn about all the progress being made with AI in radiology at Mass General Brigham and to see how AI will change the practice of radiology in the U.S. in the years to come.
Q: How is Mass General Brigham using AI in its radiology practice today?
A: At Mass General Brigham, we've made significant investments to support the creation and adoption of AI that are now bearing fruit, including more than $1 billion in our EHR and multiple decades in longitudinal data assets, notes, image repositories, genomics, etc.
In 2016, we launched the Center for Clinical Data Science (CCDS), a full-sized team solely focused on creating, promoting and translating AI into tools that will enhance clinical outcomes, improve efficiency and enhance patient-focused care. We also created what was, at the time, the largest GPU supercomputer ever deployed at an academic medical center to help process the vast amount of data we were beginning to collect.
In 2018, we announced the signing of a multi-year strategic agreement with Nuance to optimize rapid development, validation and AI utilization for radiologists at the point of care. Executed under the CCDS, the collaboration focused on improving radiologists' efficiency and report quality via algorithms that would be made available via the Nuance AI Marketplace, an open platform for developers, data scientists and radiologists that was specifically designed to accelerate the development, deployment and adoption of AI for medical imaging.
This is much of what we did in our early efforts around AI – build the infrastructure to democratize and accelerate its adoption across clinical research and the practice of radiology – defining and setting the standard of what's required for AI to be functional and add value.
We started to deploy AI in our clinical practices around the same time the COVID-19 pandemic struck. This is where our early efforts began to deliver value. Though we had started our AI research years earlier, the pandemic created a surge in use-case opportunities with the adoption of virtual visits, remote technology and a continuum of information flow that allowed us to use AI more naturally.
Today, as a result of these investments, we now have our own data sets. We've developed more than 50 algorithms for use in our clinical practice – some of which have been FDA-cleared and made available via Nuance's AI Marketplace.
One such example is the algorithm we developed for the Nuance AI Marketplace to help detect abdominal aortic aneurysms. It includes five machine learning models that run sequentially, which are more widely available to community hospitals. It quickly identifies the presence or absence of an aortic aneurysm.
It's still going through the validation process, but it will be generally available to other practicing radiologists via the Nuance AI Marketplace once cleared by the FDA. By adding it to the marketplace, the algorithms are embedded directly into the radiologist workflow using Nuance's reporting tools like Nuance PowerScribe One.
Strong collaborations with industry leaders like Nuance and the American College of Radiology have been vital in accelerating AI's adoption into radiology at scale. By combining our clinical data and machine learning algorithms with Nuance's workflow solutions and ACR's experience in standards development, we're paving the path toward clinical integration and radiology of the future.
Q: How will Mass General Brigham's radiology AI strategy evolve over the next few years?
A: AI will become more mainstream in clinical care over the next few years, and it will become an essential part of the diagnostic care process. We also foresee AI predictions utilizing multimodal data sources to drive decisions for triage and disease management through the integration of AI within the electronic medical record.
Q: What will the future look like if we have radiologists combined with integrated digital intelligence?
A: We've come a long way from five years ago when some predicted AI would replace radiologists. Instead, we see AI as augmenting the radiologist's intelligence – automating redundancies and optimizing the way radiologists practice. Not just saving time, but enhancing the diagnosis and potentially preventing what could have been an easy miss will also be critical.
With intelligent workflow, radiologists can practice at the top of their license with maximum efficiency, accelerating their ability to deliver optimal value and enable the best patient care possible.
Q: How will the emerging technology of AI transform everyday practice across healthcare?
A: A 2020 study from the American College of Radiology on radiologist uptake of AI shows that clinical adoption of AI has increased dramatically over the last five years, with 30% of radiologists indicating that they use AI in some capacity – up from none five years ago.
Over the next 10-15 years, we'll see more models become widely available and adopted, with the average radiologists practicing with 20-40 algorithms each depending on their subspecialties. These models will be better able to detect and identify rapidly declining disease states, quantify lesions on previous and current scans, and predict morbidity and mortality from a series of images.
AI can solve some of our most complex and critical health issues. For example, one area ripe for improvement is stroke care. Strokes are the leading cause of long-term preventable disability and cost $100 billion in the U.S. alone.
An MRI can detect if a patient would benefit from a procedure to remove a blood clot from a blood vessel, but most community hospitals where care is taking place don't have expensive MRI scanners. However, if community hospitals had access to AI to read CT scans better, they could better identify which patients to send for treatment.