Subscribe
Share
Share
Embed
The PancChat Podcast is a collaborative effort from Let’s Win Pancreatic Cancer and the Pancreatic Cancer Action Network (PanCAN), inspired by the long-running #PancChat Twitter/X chat.
Hosted by award-winning journalist Alisyn Camerota, each episode features conversations with leading researchers, clinicians, patients, and advocates who are shaping the future of pancreatic cancer care and research. Together, we deliver expert insights, personal journeys, and the latest breakthroughs—bridging the gap between science and lived experience.
Whether you’re a patient, caregiver, healthcare professional, or simply want to learn more, join us to connect, be inspired, and learn how you can help to accelerate progress in the fight against pancreatic cancer.
Julie Fleshman: Hi. I'm Julie Fleshman, president and CEO of PanCAN. On today's podcast, we will learn about AI and early detection of pancreatic cancer.
Alisyn Camerota: Hi, everyone. I'm Alisyn Camerota. I want to welcome our listeners to the 22nd episode of PancChat. We also want to thank our sponsor, Revolution Medicines. Now, in the last episode, you'll remember that we explored the options for pancreatic screening and surveillance.
Today, we're going to shift our focus to learning how AI can be used for early detection. Our guest is Dr. Elliot Fishman, professor in the Johns Hopkins Medicine Departments of Radiology and Radiological Science, Oncology, and Surgery. He's a member of the Johns Hopkins Kimmel Cancer Center and a co-principal investigator of the FELIX Project for Early Detection of Pancreatic Cancer.
Thanks for being here, Dr. Fishman.
Dr. Elliot Fishman: Oh, it's a pleasure. It's an honor.
Alisyn Camerota: Great to have you here. So, as we all know, early detection is vital, particularly in pancreatic cancer, yet it is very elusive, very hard to find pancreatic cancer early. And so it sounds like with the advent of new technologies like AI, that could be a huge breakthrough. So let's just start there.
I mean, in terms of AI, is that a game-changer in your mind?
Dr. Elliot Fishman: Yeah. If you think about it, you know, over the years, we've gotten better and better on CT. We got higher resolution, faster scanning, lots of knowledge that we've picked up over time. But still, if you talk about small tumors, which are, let's say, two centimeters, which is about an inch, many people have shown that 40% of them are really on the scan, are missed by the reader, even the best readers. When you look at pancreatic cancer, and I'm sure it was discussed in your last episode, we talk about survival is 13% at five years.
If you pick up small tumors, most of them are stage I, at best stage II, survival there approaches 50 to 60%. If we only picked up those small tumors, we can save 20,000 lives a year just by picking things up early. Now, you could say, why doesn't the radiologist pick them up? Well, because you're looking at a lot of things on a scan and not every scan comes in rule out pancreatic cancer. In the U.S., for example, there's 50 million scans a year done.
AI on the other hand, is the perfect scenario, is the perfect win. If we could train the computer to see those small tumors, AI can look at every case regardless of why you did the case and look for those small tumors. It can pick them up and point it out to the radiologist, here's the tumor, and then you could take care of it.
Alisyn Camerota: That is a game changer. Have we trained AI to look for the small tumors already?
Dr. Elliot Fishman: Well, you mentioned we're doing research. We've been funded by the Lustgarten Foundation for the past six or seven years. We've been working really hard on that. We actually work also with Microsoft. I have no conflict of interest.
Microsoft has something called AI for Good, so it's a philanthropy. They don't give you any money; they give you people. We've been working with them really carefully. We have a paper submitted now, which hopefully will be accepted, where we looked at tumors that were two centimeters or less. The accuracy of the computer was over 90%.
You think about it, we talk about a radiologist would be 60%, 70%, the computer would be 93%, 94%. That would make just a tremendous difference. Now, you think about, you read the papers every day, whether it's in breast cancer, whether it's in pulmonary emboli, a range of things, AI is beginning to be part of the story. It's not something that, oh, AI is going to be good five years from now. That's always what we always think about with most things.
Well, it's going to be good. Then you ask when, in five years. Now, you read about breast cancer. Everyone would say, there was an article by Eric Topol last week that made the point that if you're a woman getting a mammogram, you should make sure you get AI. It increases accuracy by 30%.
Not in five years, it's today. What we're working on is the ability to now look at all pancreatic tumors, whether you're looking for them or not, have AI look at the scan and be able to tell you which ones are abnormal. There was an article I'm sure everyone has read or heard about a couple of weeks ago where there was an executive in New York for one of the big health companies who said, I'm going to replace all my radiologists with AI. They said, when are you going to do that? He goes, well, whenever it's possible.
Maybe he doesn't like radiologists, but we're talking about that this is something that can be done hopefully very soon. Our goal is not to write papers, which we do write a lot of. Our goal is to go from bench to bedside. We need to have this out there now that it's routinely used. There's 60,000 pancreatic cancers a year in the United States.
That's a big number. Even though it's not the most common cancer, there are new drugs coming along. You mentioned Revolution Medicines. There was some really exciting news presented last week or two weeks ago at the AACR meeting. But the first thing is you've got to detect it, and you've got to detect it early.
Pancreatic cancer, the way to cure pancreatic cancer is surgery. At the end of the day, 85% of people who present with pancreatic cancer are not surgical candidates. What if we can change that number to say, instead of 15% are surgical candidates, what if we could say that 50% were surgical candidates? That would be a game-changer. Yes, we give chemotherapy for patients now before they get operated on, but it's really that first scan that tells the story of how the patient's going to do.
I think it's very exciting.
Alisyn Camerota: It is very exciting, doctor, but it also sounds like you've already proven that this — I mean, your research shows that AI can detect better than radiologists. So why isn't this — when can we scale this up? Why isn't this already ubiquitous?
Dr. Elliot Fishman: Yeah. So one of the things you have to do, so we've looked at in training and testing, we've looked at 8,000 cases of proven cancer. One of the challenges is in getting this everywhere; need to — for example, there are many manufacturers, Siemens, GE, Philips, Toshiba, and other companies that make scanners. Often, how the images look and how they're generated by different scanners will be variable. One of the big things and one of the challenges people have is when you do a study, let's say you do it at Hopkins or NYU or Mayo or wherever, you have your own material.
That may be good, but it doesn't have all the different scanners. You need to really make certain that you have a lot of material, that you're able to look at every scanner, and so wherever a person is scanned, you'd be able to make that same jump. We are working on that. Our training sets, it's very important that you don't have specific types of data only. I mean, it's very easy for me.
I'm at Hopkins to get Hopkins cases. I have an IRB. I go into the file and I can get them. But if you say to me, what about the University of Maryland, which is about two miles away? It's not a trivial point to get those scans.
One of them is the logistics of doing things, but I think we do need to be smart. I think we talk about China a lot. If you look at China, to their credit, and countries in Europe as well, the medical system is one piece. In China, they can get 30,000 scans in a day. It would take me years to get that many.
I think we need to really make access better. We're trying to really improve the results for everybody. Because one of the things that AI has done and will do, if you're at a place like NYU or Hopkins, you name the place — Penn, Mallinckrodt, Mayo — those are the top of the line places. Let's say you'll get better care theoretically at a place like that, people have more experience. At the end of the day, 98% of the patients live in many other areas.
It's a big country. There's 300,000,000 people plus spread around. What AI can promise, and I think the most impressive thing about AI to me, is that you provide the same level of care for everybody. It's not just, hey, I'm in New York. There's really good radiologists, really good surgeons, oncologists.
But if I'm in some little town in Oklahoma, my images are reviewed by the same software. You can really improve. That's the thing that's exciting about AI. It will raise the standard, it will raise all boats that everybody, when you look at things like physicians in general, radiologists, since my radiologists are like, I can speak about that easier. What AI does, if you're the world's best and you're the top of the mountain, it can make you a little bit better, maybe.
But if you're average, it'll make you like the world's expert. That's the thing that's exciting about AI, is that it raises all boats, and that you can become an expert in something that you're not really an expert, but it will change your accuracy. I think that's what's really exciting because it'll improve things for everybody, not for a select few. People sometimes forget that. That's why AI it's great for across humanity.
Alisyn Camerota: Well, that's a very heartening picture of AI because obviously, often we hear the perils of it. It's really nice to hear the upside of what it will be able to do. But what's the timeframe? Since you said that everything is always five years out, when do you see it as being used at a much bigger scale like that?
Dr. Elliot Fishman: Well, we have these really good results, and we're working hard. And the people who are funding us, I had to promise that this is not just something we're doing to write an article and do more work. We want to get that out as soon as possible. I think we have really good results. We'll do a little bit, make the process a little simpler and then get it out to other hospitals to test it on their data.
We work with other places. We work with Pittsburgh, we do some work with NYU. We have tested it at other places, but I think what you need to do is make it available to other places and let them use it. Because unless you really test it, when you're sitting there, and you're the developer, you could get some results. But if I move it 500 miles away, put it on their system and say, use it, that's really the true test of how good something is.
Our goal is to get it out as soon as possible. It could take five years, but I would be very disappointed. Our goal is to get it out far sooner than that. One of the things I promised, and when we speak our group together, it's kind of like, I don't know if you remember this, maybe this dates me. Remember the Blues Brothers?
Alisyn Camerota: Of course. And did act right, of course.
Dr. Elliot Fishman: Right. And so, you know, remember when they were trying to bring the band back, they said, "We're on a mission from God.” And I think that urgency, I think one of the things you don't want to be sloppy, you want to be very careful. But then there needs to be an urgency because if it's so helpful, we need it out there, not in five years. Get it out there now in a year and months, and whatever it needs to get it out there, make certain that you're able to solve those problems to make it available.
Alisyn Camerota: Beyond what you're talking about, AI could spot small tumors before a radiologist's eye can, how is AI already being used in cancer detection, or is it?
Dr. Elliot Fishman: AI is really changing the game of cancer. One thing, we talk about liquid biopsies. There's a couple of companies doing that and they're using now AI for determining on their results where the primary tumor is. We're talking about pulmonary emboli, which is a complication often of therapy. We're using AI now.
It's in routine practice for increasing the accuracy for detecting pulmonary emboli. In mammography, it's routine now. In Europe, it used to be till about a year and a half ago, they required two radiologists to read a scan. Now it's one radiologist and AI. If you look at all the data, there's a couple of big data sets, 500,000 plus patients.
The accuracy of AI is 30% better than the best radiologists. I mentioned that article by Topol, I read it last weekend, it came out last Saturday, which basically said that if you're really talking to patients in a sense, when you get your mammogram, make sure you get your radiologist report and you get the AI report. Now, I think you have to have the radiologists work with the AI obviously, but that's there now. That's not something that's five years from now. I think whatever city you're in, big city, small city, people have the AI available now.
There's a number of vendors. There's a lot of work with AI in drug development. Basically, you look at every company — whether it's every company, I won't even mention names of companies, but everybody — I went to the NVIDIA meeting two weeks ago, which is all the applications from robotics down to everything else, to medicine. What they were showing was the big drug companies are all designing their drugs with AI. All the oncology drugs, which would take 10 to 15 years to get to market, will get to market in under five years. They'll be able to test them also.
Also, you can come up with designs for drugs you never thought were possible with AI. If you're looking at where AI is in medicine, it's in radiology detection of disease. People are using AI in pathology. It's more accurate than pathologists in looking at slides. In drug development, in patient management, we're working now — multidisciplinary conferences are very common in oncology.
You get together, radiology, pathology, medical oncology, surgical oncology, radiation therapy, nursing, and you discuss individual patients. Really, the patient gets the expertise of everybody. But even in the biggest institutions, that takes a lot of time. You'd like to have everybody get to a multidisciplinary conference. Reality is that only about 3% of people get multi-D conference.
One of the things we're trying to do is develop AI to create the multidisciplinary conference. All your information is pushed into the computer and the computer will come up with the ideal management for you. I think the whole spectrum is very exciting in terms of that. If you think about the amount of discovery, the amount of new information that literally is 100,000 articles published a year, there's all these new drugs, hard to keep up with the material. AI can provide that for you so that you're on top of your game, that you need to know the information.
You think about Revolution Medicines, since you mentioned it's one of your sponsors, their information, and I'm a radiologist, I mean, people are really the medical oncologists who've been looking at that drug, but I saw it online. I read about it. I know what it does. We know how it can impact. All of these new drugs, the entire process of robotics and surgery.
Mentioned the computer meeting. One of the biggest things is robotics in medicine. They were showing how you can do robotic surgery where you can remove smaller tumors with less injury to normal tissue. The idea of being able to do robotic surgery, they were showing even robotic surgery — we think of robotic surgery as I'm sitting next to the robot with my hands. They were showing robotic surgery where the surgeon may be wherever you are in, let's pick Baltimore, I'm here.
Baltimore and the patient could be in India, and doing robotic surgery remotely.
Alisyn Camerota: Wow.
Dr. Elliot Fishman: It's really the ability to totally change the way we think and make what kind of was perhaps science fiction very real. It's coming very quickly because if you start looking at medical records, we think about charts and everything else. There's companies like Abridge, which basically has changed medical records so that instead of — the biggest complaint people have is I go to the doctor, the doctor looks at the computer, I'm sitting there, he or she are typing. Well, with Abridge, what happens is it's just like you and I talking and the computer is listening. The computer gets all the information.
Then you set it up the way you want, but the person who showed me Abridge, what he does is, he's a cardiologist, he creates three reports. He creates one report that goes back to himself, which has the most detail. It creates a second report that goes to the referring physician. Since he's a cardiologist, it goes to the internist, which says, "Hey, we did this. I recommended these meds.”
The third report is for the patient. “Mr. Smith, I told you, you need to take these three meds. This is when you take them. We agreed that you're going to walk once a day.”
The thing is, when the doctor and the patient met, the doctor only looked at the patient. The doctor wasn't typing. That's all AI and that's available now at Hopkins Mayo Clinic. It's this whole agentic AI where you change everything because now, A, you can do things faster. I was told, for example, that this person sees 10 patients a day and he told me he spends three hours prepping.
Now, he has a computer giving me information, so he spends 20 minutes prepping. Each part of the puzzle from the history to the image interpretation, to selecting the medication, all of that driven by AI.
Alisyn Camerota: And is there any downside? I mean, you have spelled out all of the revolutionary ideas of how AI can make it easier for doctors and better for patients. Is there any downside right now to having AI be involved?
Dr. Elliot Fishman: Well, first of all, you have to make certain the programs you use are correctly vetted. What you want to make certain is whatever AI you use — we tell this to people who do mammography. I'm not a mammo expert, but there's many different systems. It's very important you get the AI that works really well with your system.
That becomes very important. You also want to make sure that whatever AI you have was well tested. It's very easy to come up with something that works on 50 patients, 100 patients, but what's it going to do on the next 10,000 patients? You really need to make certain that it works in your environment, that it works well for you, and that it was well tested, and that it's continuing. There was an article, a friend of mine was giving a talk about AI and some of the failures.
There was an article, one of the original articles about AI and dermatology. It showed that a dermatologist — the dermatologist you and I go to, the best dermatologist — in their career, they'll see about 15,000 or 20,000 skin lesions, roughly. This program was trained on 1.2 million. So one thing is you can't compare the experience. But on the other hand, you have to be careful.
What happened with the machine? My friend, AI people will say, AI tries to find the answer, but it may not be the correct answer. What happened was with the skin lesions, no one noticed this, but if a dermatologist looks at your hand and sees a little something funny, but they know it's nothing, they take a picture. If they think it's a problem, they take a picture, but they put a ruler on it to measure it. What the computer figured out was if it had a ruler, it was probably cancer.
Alisyn Camerota: Oh, that.
Dr. Elliot Fishman: No one cheated, but the computer figured out if it has a ruler, it's bad. Right. It's things like that. You need to make certain that things really are doing what they're supposed to do.
Alisyn Camerota: Well, you say that as though it's easy for a patient to figure out if the right AI system is being used. Obviously, the hospital system needs to do that for patients. We just need to sit tight and wait for these bugs to be worked out. Tell us about FELIX 2.0. What is that?
Dr. Elliot Fishman: The FELIX 2.0 is our project for the early detection of pancreatic cancer. I'm the PI. It involves radiologists. We have a lab with a bunch of computer scientists. We have Microsoft computer scientists.
We work with our colleagues in oncology. We work with our colleagues in surgery and in medicine. It's really a multidisciplinary group trying to detect early cancer. We have all of the data. We're developing algorithms to basically be able to detect things early and then manage patients early.
We've looked at a range of different tumors, trying to not only — remember, we talk about pancreatic cancer, but there are other types of pancreas tumors. There's neuroendocrine tumors, there's things called mucinous cystic neoplasms, serous, simple cysts, IPMNs, MCNs. There's a lot of things that you need to look at beyond pancreatic cancer. We published an article with Microsoft about a year and a half ago, I think now, where there was a study done, multiple institutions, I think like 16 or 18 institutions around the world, top places looking at cystic pancreatic lesions. One of the big challenges in the pancreas, you get these cystic lesions.
97% of the time, you live your life, they're benign. Three percent, they turn malignant. The question is, how do you manage them? Who do you follow? What do you take out?
It's not that simple. You can sample fluid. There's a lot of work being done there, but sampling fluid is EUS, endoscopy, a tube going down, all sorts of complications. It ended up when you looked at this — 860 patients, they were all recommended, and all went to surgery on these cystic lesions. That means the surgeon, the GI docs, all thought it was highly suspicious for malignancy.
Well, they were resected. Remember, cancer — pancreatic surgery is either a Whipple procedure, remove the head of the pancreas, or a distal pancreatectomy, where you remove the tail. Pancreatic surgery has a 3 to 5% mortality rate. It's not getting a skin lesion taken off. So you don't want to do it unless you have to.
Also, you do things like Whipple procedure. It removes the pancreas, duodenum. It changes your absorption. It's a life changer to get that. Need it, it's a lifesaver.
But how do you determine it? It ended up that out of these 860 or so patients, only 40% of the patients actually had tumors that were malignant. 60% of the patients had surgery and really didn't need surgery. So, there was some work done by Bert Vogelstein at Hopkins, and they wrote a paper a couple of years later with some early AI, and they were able to improve from 40 to 60%, which is significant, looking at the same data. We went back, looked at the same data, added nothing new, did not reinterpret the data, but we used the newest AI to analyze the data.
And our accuracy there was, instead of being 40% or 60%, was 92%. What that meant was, for example, one little number — 120 patients who got surgery would have nothing done. They would have been followed or discharged.
Alisyn Camerota: Wow.
Dr. Elliot Fishman: So those are the things that become very, very important. That's why we're very excited. So we've been working really hard on being able to look at all the pancreatic tumors, determine what they are, detect the smallest of tumors like we mentioned, but then also to look at the aggressiveness of tumors, predict how we should manage tumors. We're also looking at pancreatic cancer as you get treatment. Can we predict how people will respond to treatment?
Can we predict who will respond to treatment? I mean, till now, pancreatic cancer was very limited. The options for chemotherapy, there are a few chemotherapy agents. Some diseases have had incredible discovery in terms of chemotherapy or immunotherapy. Pancreatic cancer hasn't changed in a long time.
You have to use what you have, but can we predict who will respond to therapy? Because the therapy typically is not very benign. Patients get really sick. If you could choose who's going to respond and what the best medication is based on just looking at the tumor, that would be very exciting. We are working on that as well.
We're trying to look at things across the spectrum, and trying to do it not being sloppy or just rushing, but trying to do it with the promise of getting it done ASAP.
Alisyn Camerota: Well, Dr. Fishman, that all sounds very exciting. I think we all understand better the complexities and just where we are right now with AI. It's very promising on the horizon, but obviously some of these kinks need to be ironed out. But thank you. Is there anything else you'd like to say that we've missed?
Dr. Elliot Fishman: No, I think the main point you made very clearly is I think patients need to be paying attention. It's a changing field. You can't say, well, what does AI say for everything? But I think people need to be paying attention. Like I mentioned, mammography.
Are supposed to get mammography every year, it's getting earlier. Well, I think right now the recommendation would be if you go somewhere for mammography, make sure they're using AI. I think it's one step at a time, but I think people will begin to feel very comfortable if they know. People at times are concerned that AI is going to take away their dealing with physicians. We don't look at it that way.
We're not looking and saying, oh, we're going to have AI and there'll be no more physicians. I think what we're saying is AI with a physician is going to improve the ability to treat you, to detect disease, to manage disease. I think that's what our goal is. Our goal is not to make ourselves unemployed, but I don't worry about that particularly. If we could — if we could come up with something that would put all the doctors on unemployment and we do all this wonderful stuff for patients, I'd be out of here today.
Alisyn Camerota: Well, that is very generous.
Dr. Elliot Fishman: Wish we could do that, but I think —
Alisyn Camerota: Hopefully, it'll be a combination of both doctors' expertise and the wonders of AI. So Dr. Fishman, thank you very much for spelling all that out, for giving us a new perspective on everything, and a great tip about mammograms. So wonderful to see you. Thank you.
Dr. Elliot Fishman: Same here.
Alisyn Camerota: Thank you to everyone for listening. I'm Alisyn Camerota, and I'll see you next time on PancChat.
Cindy Gavin: Thank you, Dr. Fishman and Alisyn, for that wonderful conversation. I'm Cindy Gavin, CEO and co-founder of Let's Win. If you or a loved one has been diagnosed with pancreatic cancer, navigating this journey can be very overwhelming. But you don't have to do it alone.
Be sure to explore the many available resources for patients and caregivers through Let's Win and PanCAN. PanCAN can be found at pancan.org, and Let's Win can be found at letswinpc.org. Our platforms continuously post news about emerging technologies and the latest treatments. Together, Let's Win and PanCAN are committed to guiding you through every step of the pancreatic cancer journey, offering support, information, and much needed hope. In our next episode, we will be speaking with pancreatic cancer survivors about survivorship.
Follow PancChat to get your latest episodes each month right in your podcast feed. PancChat is available on all major platforms wherever you get your podcasts. We look forward to seeing you next month.