William E. Karnes, MD, specializing in gastroenterology and internal medicine, clinical professor of medicine at UCI, talks about artificial intelligence technology (AI) being tested and used to help locate polyps during a colonoscopy, in turn possibly helping to prevent future cases of colon cancer.
Interview conducted by Ivanhoe Broadcast News in October 2018.
The AI is pretty cool; tell us what kind of a difference it’s making in your practice.
Dr. Karnes: We are still in the development phase but we are rapidly approaching our dream that AI will be our expert observer and recorder, helping all colonoscopists achieve and report the highest possible quality. Colonoscopy quality is ultimately measured by our ability to prevent colorectal cancer, which requires that we find and remove all precancerous polyps. Fifty percent of the screening age population have precancerous polyps. Yet only the best colonoscopists routinely find that many polyps. Most colonoscopists find these polyps 25-30% of the time, and some as low as 7-10% of the time. Missed polyps are responsible for most colorectal cancers that develop between recommended colonoscopies. Medicare recognizes this difference in quality and now requires that we report our precancerous polyp detection rates, as well as other measures of colonoscopy quality. Medicare will financially reward the best performers and penalize the lowest performers. So we created multiple AIs to record all our quality measures and to show us when a polyp is present. This will make make our lives a lot easier for recording and reporting our quality to Medicare. We hope that AI will also help all colonoscopists achieve higher detection rates of precancerous polyps and prevent up to 90% of colorectal cancers with colonoscopy. We won’t know for sure until randomized multicenter studies are completed.
So they’ve been screened in between the five years, something that was missed the first time?
Dr. Karnes: Yes. Interval colorectal cancers are those that occur between recommended screening. Most are caused by missed precancerous polyps or cancers.
How does the AI work?
Dr. Karnes: In technical terms, it’s a convolutional neural network model. It’s really good at facial recognition, things like that. So it looks at images and if you teach it with enough images it can tell the difference between a dog and a cat etc. Instead of cats and dogs, we trained the neural network on colon images with and without polyps, and we painstakingly drew little boxes around all those polyps on tens of thousands of images. The trained AI software was then tested on its ability to find polyps on a new set of images. Once it reached about ninety nine point five percent accuracy we were starting to feel confident. The next challenge was making it run fast enough. It has to look at each frame of a sixty frame per second high definition video, process that image, look for a polyp, draw a picture on it and put it on the video screen in real time. We successfully met that major challenge just over the last couple of months.
I saw little boxes. Is it in trial phase right now?
Dr. Karnes: We are in a trial phase to optimize the interface and train it to discover particularly tricky polyps th. We don’t want the AI software to miss anything, but we also do not want too many false positive results.
That’s why you have the control and the AI version.
Dr. Karnes: Right. The first challenge was to get it to run in real time, the next was to really smooth it out so we didn’t have any flicker or frame drop on the video. Then we fine tuned the interface just the way we all like it. The next step is to test it prospectively, how well is it improving our polyp detection rate, how well is it accurately reporting our quality measures, such as withdrawal time, insertion time, prep scores, and cecal intubation rate. These will require prospective multicenter studies. We’ll get lots of places around the country involved in this.
So how is the trial part working now, is it volunteer?
Dr. Karnes: Right now we just tell patients when they come in that we’ll have this extra screen and they’re usually quite wild about it and want to watch that screen instead of the other screen. But yeah they’re just volunteers to let us fine tune the interface.
How many volunteers have you had so far?
Dr. Karnes: Well for a while we had one room running the AI software just on Friday’s. That was maybe ten patients per week.
Ten for a week?
Dr. Karnes: Ten for a week and now we’re running it in two rooms simultaneously on Wednesdays and Fridays and that’s coming up to maybe fifteen to twenty per week. We’ll get them into every room eventually.
So you’ve done hundreds of patients already?
Dr. Karnes: We are probably close to a hundred live cases so far.
We talked to John Gifford this morning, one of your patients. What he thought was really cool was not being completely under, they’re sedated and they can chat with you, that’s different too.
Dr. Karnes: Yeah, yeah and that’s independent of AI. You don’t need AI to do a colonoscopy un-sedated.
Is that standard?
Dr. Karnes: Every colonoscopist is different, every patient is different on what they prefer. I offer it to everyone, about a third of my patients say, sure let’s go for it. Ninety percent of those who start without sedation remain comfortable, engaged and never ask for sedation during the exam, and they say hey, that’s the only way to go.
So the only difference from the detection of the AI, the rest of the colonoscopy is pretty standard?
Dr. Karnes. Everything else is standard. If you’re not doing a good job as a colonoscopist driving your scope and looking behind all the corners AI is not going to help you. AI can only see what’s on the image so you’ve got to get the entire surface of the colon in the image. It doesn’t affect what we’re required to do as colonoscopists except hopefully it will help us all be better and reduce the amount of redundant reporting we have to do after the procedure is finished. That gives us more time to discuss the results with our patients.
How much do you think does AI improve the detection rate?
Dr. Karnes: We don’t know yet. We did a video study in which we asked 3 expert colonoscopists with very high polyp detection rates to identify all polyps. Twenty percent more polyps were identified with the AI running compared to without the AI running. But we don’t know if the extra polyps we precancerous or just benign bumps. The results are promising, but we don’t know yet what real life is going to be.
Is it fair to say it could be as much as twenty percent?
Dr. Karnes: I think we have to be very careful about making any claims.
Let’s talk again about the twenty percent and you were talking about a goal.
Dr. Karnes. My goal is that every colonoscopist will have a detection rate of close to fifty percent for these precancerous adenomas. Finding more polyps is the key to preventing colon cancer with colonoscopy. Our video study suggests that AI will help colonoscopists find more polyps but we won’t know for sure until the randomized multicenter studies are complete.
It must be super exciting to see then this may close the gap, this AI.
Dr. Karnes: We hope so!
I read in the news report it was like having ninety or ninety-eight pairs of eyes looking instead of just two?
Dr. Karnes: I guess the AI software is another set of eyes. It was trained by expert colonoscopists who know how to find polyps well, who have high polyp detection rates. So hopefully AI will be like have a seasoned veteran standing with you, an expert who is really good at finding polyps and helping pointing them out to you. But the AI software continues to learn, continues to get better, and never gets tired.
What else haven’t I asked you that you think is important about the AI?
Dr. Karnes: It’s just exciting that it can be done in real time and it’s ready to test.
END OF INTERVIEW
This information is intended for additional research purposes only. It is not to be used as a prescription or advice from Ivanhoe Broadcast News, Inc. or any medical professional interviewed. Ivanhoe Broadcast News, Inc. assumes no responsibility for the depth or accuracy of physician statements. Procedures or medicines apply to different people and medical factors; always consult your physician on medical matters.
If you would like more information, please contact:
James Chisum
714-496-0541
jamesc@millergeer.com
Sign up for a free weekly email on Medical Breakthroughs called First to Know by clicking here.
