Using AI for Early Detection of Pancreatic Cancer? Five Questions with Dr. John Chabot

Blurred photo of a surgeon holding a digital circuit board with 'AI' and organ icons floating alongside

Every day Artificial Intelligence (AI) is advancing its knowledge of our world, and few fields are seeing such rapid progression as science and medicine. 2023 marked the year that researchers were able to test AI’s ability to diagnose pancreatic cancer based on medical records data. The initial findings were rather extraordinary and demonstrate that AI-based screening could expedite the diagnosis of pancreatic cancer.

We spoke with John A. Chabot, MD, Chief of GI/Endocrine Surgery and Executive Director of The Pancreas Center about this initial study and where we go from here.

Will you give us an overview of the findings; how AI can detect pancreatic cancer?

It’s an incredibly exciting development. The investigators simply used diagnosis codes from Electronic Health Records [EHR]. So, they did it with the Danish National Health System, and essentially, they gave the AI system all these health records of people that developed pancreatic cancer.

Eventually, when the system did its artificial intelligence thing, it was able to identify a relatively concentrated group of patients. Not all of whom developed pancreatic cancer, but a much higher percentage than when looking at the general population, by a long shot. Three years before they actually developed pancreatic cancer. Three years.

So, the investigators then wanted to look at a different population, and they used the Veterans Administration Electronic Health Record, which is a very robust EHR, and has been for a long time. They told the AI system, okay, go out now and look at this group of people. And it was able to do a similar analysis and come to similar identification of people who were at much higher risk of developing pancreatic cancer, three years before they did.

Taking this information on cancer risk, how would you know who to intervene on?

Right. This identification in and of itself is incredible. But three years before, you're not going to see anything on any of our testing.

In an unrelated study, radiologists doing artificial intelligence research, plugged in just plain old CT scans of people who ultimately did develop pancreatic cancer. And after the learning process, the system was able to identify very subtle CAT scan findings much earlier than the typical thing we're looking for to identify pancreatic cancer.

Wow. Is that because AI can identify much more nuance in a scan beyond what a human can discern?

Exactly. And collect it then validate those with thousands and thousands of cases. So, you put those two publications together, and suddenly, we're at a potential point where we could easily be diagnosing people much, much earlier than we do today.

Everybody who talks about AI in medicine makes similar points. There are lots of regulatory issues to get through. There are lots of questions like who owns all this data? A lot of social and political and regulatory issues to get through, but the science is telling us that this could be revolutionary within a year or two.

Looking forward, what would be the treatment plan for someone who was identified three years prior to getting pancreatic cancer? Is it about preventative measures of some kind?

We don’t have anything to do that right now. If we had our existing technology, we would take the area out and surgically remove it because AI can identify where the cancer would grow in the pancreas.

The AI systems are fed the images, and they have figured out what to identify as precursor observations that consistently come true. Now, how consistent is that going to be? What's the error rate going to be? We don't know any of that. But these are just incredibly exciting developments that give us, I think, a huge opportunity.

Are there wider applications of these findings in other fields of medicine?

Oh certainly. For example, in vitro fertilization: the AI is looking at the eggs and the embryos, and AI is now picking the ones that are going to be successful, not doctors looking in microscopes. The difference in take rate is phenomenal.

I think this is going to be applied across all of medicine. In my world, there just happened to be two important publications this spring, to tell us that this is coming soon, as long as we can get our act together to figure out the human side of it, rather than the science side of it.

There’s more! Read an in-depth look at how AI detects pancreatic cancer with diagnosis codes and biomarkers in this interview with Dr. Chabot.


Don’t miss an update on news and content from Healthpoints, the blog from the Department of Surgery at Columbia