Do a quick search for “puppies vs. muffins.” You’ll see a grid image of Chihuahuas and blueberry muffins. Scan through each image, and you’ll notice you’re sometimes second guessing which is which.
Our brains can immediately register the characteristics (tan color with dark spots). But because the images are so similar, it can be difficult to immediately ascertain the correct answer.
Now imagine you have 10 minutes to identify 1,000 of them. How many would you get right? And what does this have to do with health care? First we need to talk about heuristics and error rates.
A heuristic method is simply a mental shortcut. It allows people to solve problems and make judgments quickly and efficiently. We use these mental rules of thumb thousands of times a day.
For example, most people don’t have to walk into a room and decide what a chair is. When our brains see an object that has four legs and back support, we immediately identify it as a chair.
Our brains do this millions of times a day. If we weren’t able to draw instant connections like this, we would be paralyzed by the constant sensory analysis.
Humans vs. computers
If you were given 10 minutes to identify 1000 images and decide “puppy or muffin?”, the classic human error rate would likely appear. This means that on average, humans would get the answer right 95% of the time and get it wrong 5% of the time.
Computers historically lagged humans in this ability. But after several recent developments in deep learning, computers have closed the gap. A long-term study on image classification ran for years to see if computers could be trained to do better than humans. In 2017, computers achieved an error rate of just 2.3%.
The computer error rate has continued to progress ever since, thanks to rapid advancements in machine-learning and deep-learning capabilities. So what does this have to do with health care?
How this makes health care better
Let’s say you have to go to the hospital and your doctor orders an MRI. (That’s short for “magnetic resonance imaging,” meaning high-resolution pictures of the inside of your body).
An extremely well-trained radiologist can look at the images and recognize patterns the average person can’t. They use their education and experience (including heuristics) to make a diagnosis — for example, “cancer” or “not cancer.” This happens all the time in health care.
In 2017 there were 36 million MRI scans done in the United States. And there are many other types of images done each year, including X-rays, CT scans and ultrasounds. Analyzing these images accurately is vital to diagnosing a patient correctly.
Technology improves lives
Humans can be inconsistent from one day to the next, even in the best situations, with the best training, working on a full night’s sleep, and in peak physical and mental health. This is where computers (with machine learning and deep learning) can help make doctors’ jobs easier and their patients’ lives better.
When health professionals are equipped with the latest technologies, they can be freed up to give their patients more face time and more accurate diagnoses. Optum is already applying emerging technologies like machine learning and deep learning to help increase the effectiveness of diagnosis, treatment, and management of disease. And we’ll continue to push the limits of technology to make health care better for everyone.
Visit optum.com/technology to learn how Optum is advancing technology for human potential.
About the author:
Director and Chief of Staff of the Advanced Technology Collaborative (ATC), Optum
In addition to management responsibilities within ATC, Charles drives efforts around Advanced Technology Immersion Sessions. These sessions serve to expose executives and business stakeholders to a meaningful, contextualized overview of how advanced and emerging technologies can help transform their business. You can read Charles Schaller’s full profile on our People page.