Radiologists often have more cases than they can get to quickly—but some cases are more critical than others. How can they prioritize? As artificial intelligence promises to enhance the capabilities of radiologists, Dr. Kadish caught up with Dr. Orit Wimpfheimer, a consultant at Zebra Medical Vision in Israel where leading-edge tools and algorithms are making a difference.
Dr. Kadish: I’ve hear that you are doing exciting work with Zebra Medical Vision.
Dr. Wimpfheimer: I came on as a consultant to Zebra to help with their clinical perspective in creating algorithms for radiology.
Dr. Kadish: You’re originally from the United States.
Dr. Wimpfheimer: I went to school in the United States and moved to Israel right after my fellowship and started a tele-radiology company with my partner. We’ve had this company for 17 years and offer radiology services for hospitals on the East Coast. Every now and then I like to find a new challenge and we were at the forefront of the last radiology revolution which was tele-radiology about 15 to 20 years ago. That was the hot topic then. I wanted to be involved with the new disruption of the radiology market, which is artificial intelligence. I scouted out the market in Israel and offered my services to Zebra. We subsequently created a relationship where, as clinical director, I bring the clinical perspective, allowing them to understand radiologists and guide them to chose algorithms that will be most useful to radiologists.
Zebra is approximately five years old. It’s based here in Israel and it has amassed massive amounts of data from patient imaging—both patients in the United States and here in Israel. With all that data and high-quality computer engineers, Zebra started creating algorithms. It’s a very new concept in general—just a few years old years old. There’s a lot of companies out there trying to work with it. They have three FDA-patented algorithms out and they are creating a many more to assist radiologists and not just radiologists because often radiologists are not available to read the studies timely, so it will assist other doctors as well.
Dr. Kadish: What kinds of x-rays do you see this kind of technology being the most accurate for?
Dr. Wimpfheimer: There’s a great range. It’s extremely helpful for emergent cases such as intracranial hemorrhages and pulmonary embolisms. They improve both accuracy and speed. Radiologists are extremely overworked and in many setting they cannot get to the cases fast enough, so this allows them to prioritize cases, allowing them to first get to the cases that have positive studies by the algorithm and leave the ones that are not positive for later. In this way the patient with the real need gets attention faster. That’s one version.
Dr. Kadish: You said that accuracy is improved. So is there data for Zebra or other technology that shows that it’s better than a radiologist or better than a physician interpreting films?
Dr. Wimpfheimer: The more sensitive that you create the algorithm, the more false positives you’re going to get as a results. There’s always a delicate balance in being more sensitive or more specific for radiologists. In an imaging center that has very high-quality radiologists, the results are about the same. But the added benefit of triaging the positive cases coupled with a second opinion of the algorithm, has the potential to create a higher quality radiology work output.
Dr. Kadish: When you say “in order to be more accurate,” what you mean is in order to be more sensitive, you have to create more false positives.
Dr. Wimpfheimer: Correct. But that’s just one way for algorithms to help radiologists. The other, which Zebra is really good at, is screening populations for findings that radiologists tend not to focus on in order to find common health problems. We have an algorithm now that’s already being used very widely in Great Britain where we find vertebral body compression fractures, which are a sign of osteoporosis. Radiologists don’t always focus on these chronic fractures when patients come in for unrelated symptoms. But if we can highlight those patients through an algorithm that screens all the chest and abdomen CT scans, we can get these people into a treatment plan for osteoporosis when they didn’t even know they were affected. With treatment, we can prevent future osteoporosis fractures such as hip fractures and decrease the morbidity and mortality associated with such fractures.
Dr. Kadish: How do you see the technology evolving in the next few years?
Dr. Wimpfheimer: It’s evolving toward increasing accuracy, increasing speed, increasing patient populations’ health…but for now, we are focusing on creating more and more algorithms because the whole technology is still relatively new. Currently we are focusing on assisting the radiologist to be more accurate and more efficient. In the future, ultimately we may be able to replace the radiologist, but we are still quite a few years away from that.
Dr. Kadish: Obviously there’s a cost to Zebra’s product or any other AI product that interprets x-rays, and presumably in an HMO or full-risk model, one can think about how to incorporate it, but in a fee-for-service model, do you anticipate any barriers to implementation because there’s no additional reimbursement for the AI model?
Dr. Wimpfheimer: That’s what’s amazing about Zebra’s model because it has an “all in one” business model. For a small fixed fee, the hospital can benefit from all of Zebra’s algorithms that are currently FDA approved and all future algorithm. Zebra algorithms should increased radiology speed, efficiency and accuracy and as such hospitals will see the need for our product.
Dr. Kadish: What challenges do you see in the increasing use of AI in radiology?
Dr. Wimpfheimer: The challenges we’re facing now are juggling between sensitivity and specificity so as to be able to assist radiologist without creating many false positive which will hinder the radiologist workflow. Also, we need to create an atmosphere that assists radiologists without having them feel threatened. From what I can tell right now, those are the two main challenges. Once radiologists get comfortable with AI—and we still have some time to go before that happens—I think it will be a huge asset to each radiology department. We have multiple hospitals that are already using the current algorithms that we have and we’re getting positive feedback. We’re always fine tuning the algorithms and the solutions we provide.
Dr. Kadish: Has it reached the point where you’ve incorporated AI into your tele-radiology practice?
Dr. Wimpfheimer: I’ve been at Zebra for only a few months, so it has not. Eventually, I think all tele-radiology practices will have some sort of AI. It’s a new phenomena and I don’t think anyone has incorporated it yet, but everyone is talking about it and it’s the hottest topic at every conference. We’re still on the brink of that happening but it hasn’t happened yet.