Radiology Vs AI

radiology vs AI.jpg

It’s the age-old story of human labor vs. machine. The assembly line worker replaced by robotic machines. Cashiers traded out for self-service technology. Taxi drivers usurped by apps like Uber and Lyft.

Even the field of medicine has a history of battling between embracing advances in technology and reserving tasks and job descriptions for human staff. Radiologists are among those who are contending with technological advancement.

One the one hand, technology can allow for faster reads and alleviate an already heavy workload. On the other hand, some things cannot be programmed. Such as a complex diagnosis or face-to-face explanation of a condition.

In this article we will look at the rising tide of AI and what it means for radiology and telemedicine in the future.

Benefit of AI: Increasing the Speed of Play and Improving Efficiency

The doctors, CEOs and administrators of radiology groups that we have talked to all say the same thing: AI is definitely going to change and create efficiencies for the physicians. But they don’t see it ever replacing a physician.

There’re too many factors with diagnostics and the human body that are fluid and that need to be considered before by a physician before an interpretation is made.

But your bread and butter studies might work really well with an algorithm. Is there a fracture on this femur? But will it be able to say everything about the fracture? Maybe not, depending on the severity.

There’s certainly no doubt that it will improve the speed of play.

In a 2017 NPR article on the subject, Dr. Bob Wachter, author of The Digital Doctor says, "Radiology, at its core, is now a human being, based on learning and his or her own experience, looking at a collection of digital dots and a digital pattern and saying 'That pattern looks like cancer or looks like tuberculosis or looks like pneumonia. Computers are awfully good at seeing patterns."

real rads.jpg

For example, AI has already been able to identify a brain bleed or not. The computer actually knows enough about the anatomy of the brain and, using a CT scan, it can create a 3d image and determine if there is a bleed. This takes about a minute to complete. Once this algorithm has run and determined there is no brain bleed at 99.99%, then it can rule out that the scan needs to be read as a critical. So, on our end as a teleradiology company, we know based on this algorithm whether a scan needs to be read in 15 minutes or less (critical) or 30 minutes or less (STAT).

This makes sure that injuries with brain bleed, or similar critical issues, are automatically pushed to the top of the list.

Some radiologists are excited about the future of AI as it relates to prioritizing scans and creating better efficiencies. They foresee their jobs as less time spent analyzing images and more time interpreting results based on selected algorithms.

What AI Can’t Do (or Replace)

Depends on the style of modality and the type of x-ray. Take a knee or ankle, for instance. It may be necessary to do a 3-view x-ray of the area. An algorithm would need to understand so many aspects for that read. An x-ray won’t show flesh very well and it won’t get into tendons or ligaments, an MRI is needed for that.

An MRI aligns your ions and then takes an image off of the RF (radio frequency). For a knee or ankle or shoulder, an MRI scan could contain a multitude of detailed images. A computer, no matter how advanced, will not be able to determine which image to focus on to deliver the best, most thorough, read.

Then when you get into digital mammography or tomosynthesis, they will have ten times the amount of studies or images. A radiologist will be able to look through these images and determine which scans to focus on. The dataset is too large for a computer algorithm to make that kind of educated decision (we’re talking more than intelligence here – professional history, experience, understanding of intricacies and human error all go into this intelligent read).

There is also interpretation and analysis when it comes to the patient worksheet that accompanies the scan. When a radiologist is presented with an x-ray or MRI scan, they also get a worksheet that explains, for example, this patient has pain lifting this shoulder or there’s a tightness when he bends a certain way. Then, the radiologist is already putting the pieces together and knows to focus on a certain ligament or tendon when reviewing the scans.

AI is great for determining the presence or absence of, as we said above, a brain bleed. But when it comes to also taking into consideration patient history, size of data set, and diagnostic interpretation – a computer is simply not going to be able to provide that depth of analysis.

What Does the Future Look Like for Radiology and AI?

It should be clear from the above section that radiologists are not at risk of being replaced by machines. The stakes are too high, and their jobs are too intricate to be boiled down to an algorithm. But, that doesn’t mean there isn’t a place for AI in the future of radiology. By increasing speed, creating automations, and bettering efficiency in the journey from scan to interpretation, AI is a useful tool that radiologists can use to help maintain quality patient care while managing large workloads.