How is Artificial Intelligence Solving Healthcare Challenges?
In the world of healthcare, information is flowing faster than ever—and artificial intelligence (AI) will be needed to realize healthcare’s big data potential.
A recent analyst report found that data related to healthcare was one of the fastest growing of any industry. It’s predicted that the total volume of medical information will rise from 153 exabytes in 2013 to 2,314 exabytes1 by 2020.
While the growing immensity may be overwhelming for an individual doctor, practice or even network, it’s good news for applying AI and machine learning which can help manage the challenging scale of this data.
But there’s a bigger problem. The bulk of that information is unstructured, or dark data. Instead of being stored in spreadsheets and databases, it’s free-form digital notes and other documents. Or worse, the information is written comments scrawled on paper or physical images, like x-rays.
What are the ways we can clean up and apply artificial intelligence to put all this healthcare data to use? We spoke to three companies who are leveraging AI in different ways to tackle the healthcare data deluge head-on.
1) Prepping for Artificial Intelligence: Adding Structure to the Unstructured
How do you begin to make sense of “messy” healthcare data? A good start might be to try to give it some structure. Ciox Health manages the health records for 60 percent of hospitals in the U.S., and is working on standardizing and indexing all the data these hospitals collect so it can be leveraged to make better business and care decisions.
Florian Quarre, Ciox’s Chief Digital Officer says the company starts any engagement with a hospital by getting records into some form of digital shape.
“Ideally we prefer direct database access,” he says, “but when that’s not available we often resort to screen scraping through optical character recognition—turning an analog scan into digital text.”
In the worst cases (think: handwritten scribbles), human workers may have to manually enter the information into the system. “Then we can apply artificial intelligence natural language processing tools to curate the information by pulling out names, addresses, and diagnoses,” he adds. By giving structure to formerly unstructured data, physicians get a much more complete look at a patient’s history and can diagnose conditions with better accuracy.
2) Taking it further: AI Mining Medical Data for Answers
While structuring data is a great start, machine learning developer DataRobot is using AI to take things a step further by searching for patterns and trends amid mountains of medical data.
“Today we can take structured and unstructured data and analyze it to build insights—insights we can use to make a proper medical prediction,” says Bill Moschella, DataRobot’s GM of Healthcare.
In medical fields such as clinical research, turning data into medical insights is a tried and true process, albeit an incredibly complex one.
“Today a researcher has to acquire sample data, import it using various tools into a digital notebook, and eventually attempt to use that data to train a model by feeding it into various algorithms,” says Moschella. “The problem is that this is a months-long, arduous process requiring a data scientist with lots of skill. You spend more time tuning the analysis than actually solving the problem.”
DataRobot makes this process faster by building hundreds of these algorithms, and then putting them to work simultaneously to determine what are the best artificial intelligence models. After it identifies the most suitable algorithmic model, DataRobot’s machine learning platform continuously refines that model as new data is received. This tool is used in a range of applications, from understanding patterns in a clinical experiment to predicting how an individual will respond to a certain treatment plan.
3) Using Artificial Intelligence & Data for Financial Decisions
In addition to improving things like diagnosis and treatment, companies like Health Catalyst are using unstructured data to improve safety in hospitals and reduce needless financial expenses for both patients and healthcare providers.
Mike Dow, Senior Director of Product Development at Health Catalyst, offers one example: Say a patient is admitted to the ER after a heart episode—what are the risks of that person being readmitted after their release? The goal is to keep patients out of the emergency room by understanding the underlying factors that feed into a readmission.
“We might find out through our machine learning algorithms that whether a patient has transportation available is a key factor in determining readmission, because they can’t keep doctor’s appointments,” Dow says.
This, in turn, might be used to convince an insurance company to pay for a (less expensive) taxi ride to the doctor to avoid a (pricey) trip back to the ER.
For any given scenario, Dow says, “we want to understand the risk, and determine at the point of care what we can do to change that risk. If you can show that the risk of readmission drops by 35 percent based on factors that you’ve unearthed, you have a great financial argument in front of you.”
Unstructured data isn’t going away any time soon: Hospitals and doctors still routinely fax one another, only adding to the glut of dark data information. But while the medical industry is slowly moving towards digital processes, AI and machine learning tools are increasingly being put to the task of making sense of this information ultimately helping patients, doctors and administrators alike.
How do you see artificial intelligence improving the medical field/use of medical data?
Let us know in the comments.
This content is produced by WIRED Brand Lab in collaboration with Western Digital Corporation.
Source: 1. EMC