Natural language processing (NLP)—the concept of training computer programs to extract specific content from words and phrases—has existed in one form or another since the 1950s, but its potential to impact radiology has only been brought into focus in recent years. Vendors all over the world, including vRad, a MEDNAX company, are working around the clock to see just how much NLP can do to revolutionize healthcare—and they are finding that it can do quite a lot. 

Benjamin W. Strong, MD, chief medical officer at vRad, is leading the company’s charge as it develops a single, all-encompassing NLP engine that can be applied to any dataset, at any facility, at any time. And though many in the medical imaging industry are more focused on other innovations—artificial intelligence (AI) in particular—Strong sees NLP as an absolute game-changer for radiologists. 

“Everybody is talking about AI right now,” Strong says. “The press is going kind of nuts over it and everyone thinks image detection is so exciting. But from the perspective of practice management, radiologist workflow and national health, NLP is going to deliver far more than image-based AI in the coming years.”

Perhaps the most immediate way NLP can help radiologists, Strong explains, is providing across-the-board consistency to radiology reports. Instead of implementing a structured report template at a facility, training the facility’s entire team of radiologists to use the template and then worrying about whether they do—or don’t—comply with that template, imaging leaders can simply apply an NLP engine such as the ones designed by vRad. 

“When a radiologist creates a draft report, we apply a NLP system that does multiple things simultaneously,” Strong says. “It formats the report and it categorizes statements by organ or structure. So, the NLP is helping create a structured report, and it does that based on a draft created by the radiologist that gives him or her flexibility. Radiologists don’t even need to use a specific template or report findings in a predetermined order. After NLP, their report will look like all the other reports in the system.”

Considering the wide variety of ways radiologists convey the same basic message, this can be a more daunting task than one might expect. For example, Strong recalls, his team once examined 600 CT angiographies of the chest for pulmonary embolism. In those 600 exams, vRad found 92 different ways radiologists indicate there is no pulmonary embolism and 48 different ways radiologists indicate there is a pulmonary embolism. Getting that engine to recognize each of these different words or phrases—known as expression recognition—is a key part of the NLP process. It’s a complex process, to be sure, but it’s also exactly the kind of problem NLP was designed to solve. 

At the end of the day, this technology allows radiologists to focus more on completing their reads and providing value instead of making them learn a brand new template from scratch. “It’s a beautiful way to enable structured reporting as opposed to forcing radiologists to follow a template and change their preferred order of dictation,” Strong says. “And when you can enable structured reporting without altering anything about an individual radiologist’s approach, you can be very successful.”

The Many Benefits of NLP

Strong notes that NLP is about much more than formatting radiology reports into a preferred format; the improved consistency also helps satisfy reimbursement criteria. Acronyms such as MACRA (the Medicare Access and CHIP Reauthorization Act) and MIPS (Merit-Based Incentive Payment System) might give billing specialists headaches, but NLP can help ensure providers don’t get shortchanged when receiving payments. 

Reports can even be tailored to the needs of a single facility or referring physician, as Strong explained with an example. “We provide NLP to a well-known pediatric oncology facility,” he recalls. “They called us one day and said their anesthesia service has asked for an exquisitely detailed description of the trachea anytime they report a pediatric chest x-ray with a hilar mass, mediastinal mass or adenopathy. So we simply built that into our system for that provider. It now automatically checks for that description of the trachea anytime one of those findings is mentioned in the radiologist’s report.” 

This customization can also help by simplifying things if, say, a referring clinician dislikes a certain term. It would be a lot for the radiologists to remember on their own, but it’s no problem when the NLP can be programmed to take certain actions when the term in question is entered into the report. “It would be too much for the radiologist remember and still perform good radiology otherwise,” Strong says. 

NLP can also be trained to immediately contact referring physicians if certain findings are indicated in a report. The radiologist does not even have to do anything—if a particular finding is listed, the system can automatically start dialing the referring physician’s number to guarantee the two parties speak as soon as possible.  

50 Million Reports and Counting: The Future of NLP at vRad

Strong is intensely proud of vRad’s progress with NLP, but he says much more is right around the corner. Once the company’s single NLP engine is complete, which could be as early as late 2018, vRad will finally unleash it on their full archive of radiology reports. 

“We have 50 million radiology reports dating back to the beginning of our existence, and the ability to use NLP to analyze that data will be an enormous step forward,” Strong says. “It will help us from an operational standpoint—tracking volume, growth and modalities—and it will help us from a resource utilization standpoint. It will also allow us to track imaging utilization over time.”

At that point, Strong adds, vRad will be able to dive so deep into the data that healthcare providers are going to be blown away. If you’re at a small rural hospital ordering a head CT for a nine-year-old girl, for instance, vRad’s NLP engine will be able to look at data from their 50 million radiology reports and tell the referring physician the likelihood of a positive result. 

“It allows you to really get a handle on resource utilization, image ordering behavior of specific physicians and image ordering efficiency for specific physicians,” Strong says. “It will really change the way people think about studies.”

Yet another way NLP is set to change the lives of radiologists, and one that will likely leave them drooling in anticipation, is that it will soon save them from studying prior reports before every read.   

“Radiologists always have to read prior reports to identify things they need to mention in the current study,” Strong says. “This can be greatly facilitated by NLP, and that’s one of the first things I’ll be turning our NLP loose on once our engine is complete. Statements from any digital documents within prior reports will be categorized as ‘positive’ or ‘negative’ and listed, with links to the specific images, in the patient’s imaging timeline so that the radiologist no longer has to open prior reports and speed-read their content.”

AI may be getting the lion’s share of the attention right now, but once imaging leaders are able to take full advantage of NLP, they’ll likely find that they have a new favorite technological breakthrough. Until then, Strong and his team at vRad are staying focused on the task at hand: helping radiologists across the country provide better patient care on a daily basis. 

Click to view a sample of vRad’s Custom Structured Report.