Keith J. Dreyer, DO, PhD, is vice chairman of Radiology and executive director of the MGH & BWH Center for Clinical Data Science. The center is a collaboration of Mass General and Brigham and Women’s Hospital. Its mission is to create, promote and commercialize artificial intelligence for health care.
What is machine learning?
It’s the way you create artificial intelligence (AI) in 2017. Instead of a human programming the computer, you give it the data and the answers and let it create the algorithm, which is a mathematical rule or procedure for solving a problem. This process, in effect, teaches the computer a new domain (such as medical imaging) so it can predict answers given new data.
The concept of using machines to create artificial intelligence has been around for more than 50 years, but now companies are investing and algorithms are getting more accurate.
Will this become a major factor in health care quickly?
A good analogy is self-driving cars. You’re not going to wake up one day and suddenly find lots of cars driving themselves. Yet, we have already begun to see car manufacturers offer such AI features as parking-assist, lane-changing and traffic light detection. Over time, more and more features will be added. As in automotive, AI in healthcare will happen slowly and relentlessly.
It might start with artificial intelligence applications that will aid physicians with certain tasks. For instance, it could help radiologists detect and measure subtle lesions on CT, MRI or PET scans that they might otherwise have missed.
Why are we talking about this now in relation to medicine?
In recent years, larger amounts of data have become available across the Internet. Computation has gotten much faster. The concept of using machines to create artificial intelligence has been around for more than 50 years, but now companies are investing and algorithms are getting more accurate. People can ask more challenging questions and try to improve the simulation of human neurons. One example is ImageNet.
What is it?
ImageNet, an ongoing project, has made millions of images available of objects such as cats, dogs, beaches and mountains. Participating companies and individuals take that data set and try to create an algorithm so that a computer can find similar objects in new images.
So that it can tell a beach from a dog, for instance?
Or that there is a beach and a dog in the same image. Humans are accurate 95 percent of the time at this. In 2014, a couple of the companies were able to use machine learning to be more accurate than humans – superhuman for the first time.
How does this relate to medicine?
ImageNet algorithms were trained to look at millions of images with cats, dogs and beaches. Well, now we need to train them to look at millions of MRI, CT and X-ray images showing things such as lung cancer, breast cancer or a hemorrhagic stroke and infarction.
Computers using artificial intelligence could, for instance, very precisely quantify pathology from images in a patient over time, kind of like what we do now with blood tests. Providers could then precisely review areas of concern, comparing them historically to see if their therapy is working.
How many images does Mass General have?
Two billion. Today, you can’t take that all on at once with artificial intelligence, you have to partition the learning challenges into domains. One of several domains we’re working on is cancer detection. It involves things like mammography, lung cancer screening, and colon screening with virtual colonography. We’re also working on domains that have strong effects on the patient’s recovery, like acute neurological problems. A patient might come to the emergency room with signs and symptoms that suggest a stroke, a hemorrhage or a brain tumor. We’re working on algorithms to automatically detect those.
Are images your only focus at the Center for Clinical Data Science?
No. We’re also looking at genetic and genomic data, laboratory data and other data available inside the electronic medical record and beyond.
So I think we’ll see more and more philanthropic interest in this area because it’s going to be such a broad provider of improvement across all of healthcare’s domains.
Is Mass General positioned to be a leader in this field?
We are well poised. Our leadership took these challenges on very early. We have incredibly large amounts of data, and we have been using electronic records for a long time. Now, we’re on a common platform, eCare, so we have all of our data aggregated in an easy way to use for these experiments. And we have thousands of researchers asking deep questions and who can actually put such tools into the workflow to change care. We also have relationships with industry so that we can integrate this into commercial solutions.
What role can philanthropy play?
This is clearly one of those foundational things, like the discovery of the microscope. So I think we’ll see more and more philanthropic interest in this area because it’s going to be such a broad provider of improvement across all of healthcare’s domains. And that philanthropy is important to our efforts because training computers to learn healthcare with AI is a series of remarkably challenging tasks requiring years of dedicated research with state-of-the-art equipment and staff. There’s a high demand for data scientists today and we need to recruit the best and the brightest.
To learn more about how you can support artificial intelligence research at Mass General, please contact us.
The Research Institute: Saving Lives Through Science
The Massachusetts General Hospital Research Institute is the largest hospital-based research program in the United States, with a community of over 10,000 people working across more than 30 institutes, centers and departments.
Our researchers work side-by-side with physicians to pioneer the latest scientific advancements for curing disease and healing patients in Boston, across the United States and around the world.
To learn more about the Research Institute, please visit our website.